Number 112 January 2018 ERCIM NEWS www.ercim.eu

Special theme

Also in this issue:

Research and Innovation: Computers that negotiate on our behalf Editorial Information Contents

ERCIM News is the magazine of ERCIM. Published quar - terly, it reports on joint actions of the ERCIM partners, and aims to reflect the contribution made by ERCIM to the European Community in Information Technology and Applied Mathematics. Through short articles and news items, it pro - vides a forum for the exchange of information between the JoINt ERCIM ACtIoNS SPECIAL tHEME institutes and also with the wider scientific community. This issue has a circulation of about 6,000 printed copies and is 4 Foreword from the President The special theme section “Quantum also available online. Computing” has been coordinated by 4 Jos Baeten, Elected President of Jop Briët (CWI) and Simon Perdrix ERCIM News is published by ERCIM EEIG ERCIM AISBL (CNRS, LORIA) BP 93, F-06902 Sophia Antipolis Cedex, France Tel: +33 4 9238 5010, E-mail: [email protected] 5 Tim Baarslag Winner of the 2017 8 Quantum Computation and Director: Philipp Hoschka, ISSN 0926-4981 Cor Baayen Young Researcher Information - Introduction to the Award Special Theme Contributions by Jop Briët (CWI) and Simon Contributions should be submitted to the local editor of your 6 ERCIM Cor Baayen Award 2017 Perdrix (CNRS, LORIA) country 7 ERCIM Established Working 10 How Classical Beings Can Test Copyrightnotice Group on Blockchain Technology Quantum Devices All authors, as identified in each article, retain copyright of their by Stacey Jeffery (CWI) work. ERCIM News is licensed under a Creative Commons 7 HORIZON 2020 Project Attribution 4.0 International License (CC-BY). Management 11 Keeping Quantum Computers Honest (or Verification of Advertising 7 ERCIM “Alain Bensoussan” Fel- Quantum Computing) For current advertising rates and conditions, see lowship Programme by Alexandru Gheorghiu http://ercim-news.ercim.eu/ or contact [email protected] (University of Edinburgh) and Elham Kashefi (University of ERCIMNewsonlineedition Edinburgh, CNRS) http://ercim-news.ercim.eu/ 13 Experimental Requirements for Nextissue Quantum Computational October 2017, Special theme: Digital Humanities Supremacy by by Alex Neville and Chris Sparrow Subscription (University of Bristol) Subscribe to ERCIM News by sending an email to [email protected] or by filling out the form at the 14 From Classical to Quantum ERCIM News website: http://ercim-news.ercim.eu/ Information – Or: When You Have Less Than No Uncertainty EditorialBoard: by Serge Fehr (CWI) Central editor: Peter Kunz, ERCIM office ([email protected]) 15 Preparing Ourselves for the Threats of the Post-Quantum Era Local Editors: by Thijs Veugen (TNO and CWI), Austria: Erwin Schoitsch ([email protected]) Thomas Attema (TNO), Maran van Cyprus: Georgia Kapitsaki ([email protected] Heesch (TNO), and Léo Ducas France: Steve Kremer ([email protected]) (CWI) Germany: Michael Krapp ([email protected]) Greece: Lida Harami ([email protected]), 17 Beyond Artemios Voyiatzis ([email protected]) Key Distribution Hungary: Andras Benczur ([email protected]) by Georgios M. Nikolopoulos Italy: Maurice ter Beek ([email protected]) (IESL-FORTH, Greece) Luxembourg: Thomas Tamisier ([email protected]) Norway: Poul Heegaard ([email protected]) 19 Quantum Lego: Graph States for Poland: Hung Son Nguyen ([email protected]) Quantum Computing and Portugal: José Borbinha, Technical University of Lisbon Information Processing ([email protected]) by Damian Markham (LIP6, CNRS Spain: Silvia Abrahão ([email protected]) - Sorbonne Université) Sweden: Kersti Hedman-Hammarström ([email protected]) Switzerland: Harry Rudin ([email protected]) The Netherlands: Annette Kik ([email protected]) W3C: Marie-Claire Forgue ([email protected])

ERCIM NEWS 112 January 2018 20 Graph Parameters and Physical RESEARCH ANd INNovAtIoN EvENtS, IN bRIEf Correlations: from Shannon to Connes, via Lovász and Tsirelson This section features news about Reports by Simone Severini (University research activities and innovative 46 4th International Indoor College London) developments from European Positioning and Indoor Navigation research institutes Competition 22 High-Speed Entanglement Sources for Photonic Quantum 36 Computers that Negotiate on Our 47 Joint 22nd International Computers Behalf Workshop on Formal Methods for by Fabian Laudenbach, Sophie by Tim Baarslag (CWI) Industrial Critical Systems and Zeiger, Bernhard Schrenk and 17th International Workshop on Hannes Hübel (AIT) 38 Faster Text Similarity Using a Automated Verification of Critical Linear-Complexity Relaxed Word Systems 23 A Formal Analysis of Quantum Mover’s Distance by Ana Cavalcanti (University of Algorithms by Kubilay Atasu, Vasileios York), Laure Petrucci (LIPN, CNRS by Benoît Valiron (LRI – Vasileiadis, Michail Vlachos (IBM & Université Paris 13) and Cristina CentraleSupelec, Univ. Paris Research – Zurich) Seceleanu (Mälardalen University) Saclay) 40 The CAPTCHA Samples Website 48 10th International Conference of 24 Diagram Transformations Give a by A lessia Amelio (University of the ERCIM Working Group on New Handle on Quantum Calabria), Darko Brodić, Sanja Computational and Circuits and Foundations Petrovska (University of Belgrade), Methodological Statistics by Aleks Kissinger (Radboud Radmila Janković (Serbian Academy University) of Sciences and Arts) Announcements 48 Visual Heritage 2018 27 Quantum Computers and Their 41 APOPSIS: A Web-based Platform Software Interfaces for the Analysis of Structured 48 IFIP Networking 2018 by Peter Mueller, Andreas Fuhrer Dialogues and Stefan Filipp (IBM Research – by Elisjana Ymeralli, Theodore 48 Web Science Conference 2018 Zurich) Patkos and Yannis Roussakis (ICS- FORTH) 49 CIDOC Conference 2018 28 Chevalley-Warning Theorem in Quantum Computing 42 Can we Trust Machine Learning 49 Postdoc Position in the Research by Gábor Ivanyos and Lajos Rónyai Results? Artificial Intelligence in Project“Approximation (MTA SZTAKI, Budapest) Safety-Critical Decision Support Algorithms, Quantum by Katharina Holzinger (SBA Information and Semidefinite 30 Hardware Shortcuts for Robust Research), Klaus Mak, (Austrian Optimization” at CWI Quantum Computing Army) Peter Kieseberg (St. Pölten by Alain Sarlette (Inria) and Pierre University of Applied Sciences), 50 ERCIM Membership Rouchon (MINES ParisTech) Andreas Holzinger (Medical University Graz, Austria) In Brief 31 Quantum Gates and Simulations 51 CWI hosts EIT Digital’s New with Strongly Interacting 44 Formal Methods for the Railway Innovation Space Rydberg Atoms Sector by David Petrosyan (IESL-FORTH) by Maurice ter Beek, Alessandro 51 CWI merges with NWO Institutes Fantechi, Alessio Ferrari, Stefania Organisation 33 Control of Quantum Systems by Gnesi (ISTI-CNR, Italy), and Broken Adiabatic Paths Riccardo Scopigno (ISMB, Italy) 51 Google Awards Grant for fake by Nicolas Augier (École News detection to FORTH and polytechnique), Ugo Boscain 45 The Impacts of Low-Quality University of Cyprus (CNRS) and Mario Sigalotti (Inria) Training Data on Information 34 The Case for Quantum Software Extraction from Clinical Reports by Harry Buhrman (CWI and by Diego Marcheggiani (University QuSoft) and Floor van de Pavert of Amsterdam) and Fabrizio (QuSoft) Sebastiani (CNR)

ERCIM NEWS 112 January 2018 3 Further, experience has taught us that it is not good to limit foreword the number of members per country to one. This led to artifi - cial consortia that fell apart at some point. The model of one from the organisation or consortium per country has failed. We can and should have more ERCIM members per country. For President instance, not only CWI could be a member in the Netherlands but also TNO. The Board of Directors should ERCIM is a great organisa - see it as a joint effort to recruit new members. tion with a lot of potential. I am proud that I am the third I want to maintain ERCIM’s achievements, namely: CWI director since • ERCIM News. This magazine is widely appreciated and is ERCIM’s foundation in becoming ever more important. 1989 who has become a • Awards. In addition to the Cor Baayen Award for the more President of ERCIM, after fundamentally orientated research, we could establish an Cor Baayen and Gerard van Oortmerssen. I will outline my innovation award, which would be a good supplement and ideas on ERCIM’s strategy below. which is in line with the increased importance of valorisa - tion. The number of ERCIM Members has declined in recent • Programs such as the ERCIM Alain Bensoussan Fellow - years. I want to focus more on our roots, what we are, and ship programme and the PhD program we are currently how we started: as an association with a focus on research working on to develop young talent. institutes in the fields of both informatics and mathematics. • The Working Groups. I want to revive the former working Of course, leading research universities are also welcome to groups by setting up new working groups on hot topics become ERCIM members, as we have university members such as blockchain. This will allow us to really benefit now. Our focus will distinguish us from other organisations. from our mutual collaboration within ERCIM by coordi -

Jos baeten Elected President of ERCIM AISbL

The General Assembly of the ERCIM AISBL, held on 24 October in Lisbon, unanimously elected Jos Baeten, gen - eral director of Centrum Wiskunde & Informatica (CWI) in the Netherlands as its new President for a period of two years as of January 2018. Jos Baeten succeeds Domenico Laforenza from the Institute for Informatics and Telematics (IIT) of the Italian National Research Council (CNR) who served as President of ERCIM since January JosBaeten(left)andDomenicoLaforenza. 2014. January 2015, he is part-time professor guests from academia, industry and Jos Baeten has a PhD in mathematics in theory of computing at the Institute politics, his efforts in approaching from the University of Minnesota of Logic, Language and Computation ERCIM and Informatics Europe to (1985). From 1991 to 2015, he was of the University of Amsterdam. He is cooperate, and his remarkably professor of computer science at the well-known as a researcher in model- inspiring speeches were highlights Technische Universiteit Eindhoven based engineering, in particular in during Domenico’s presidency”, Jos (TU/e). In addition, from 2010 to 2012 process algebra. says. Domenico began his activity in he was professor of systems engi - ERCIM in 1993, contributing to the neering at TU/e. As of October 2011, Jos Baeten expressed his gratitude to creation of the ERCIM Parallel he is the general director of Centrum Domenico for his dedication to Processing Network (PPN). He has Wiskunde & Informatica (CWI) in ERCIM during his four years of presi - represented CNR on the Board of Amsterdam, the national research dency. “The festive 25th anniversary Directors since 2006 and will continue institute for mathematics and computer celebration of ERCIM at CNR, where representing CNR after his presidency. science in the Netherlands. Since Domenico welcomed over a hundred

4 ERCIM NEWS 112 January 2018 Joint ERCIM Actions

nating research in those new fields. Working groups have a limited life span, focusing on temporary, topical, short- tim baarslag Winner term research that responds to new and current develop - ments. In terms of collaboration, it’s important that of the 2017 Cor baayen researchers, not just directors, will be able to find each other. This is also the idea behind exchanging researchers Young Researcher Award in our Fellowships and PhD programme. Added value can be found in new research areas. Tim Baarslag from CWI was selected as the winner of the 2017 ERCIM Cor Baayen Award, in a very tough ERCIM also successfully coordinates EU projects. This task competition with 15 finalists. The award committee is carried out by the EEIG office, originally conceived as a recognises Tim’s skills and the results that he has service for the members. I will strongly encourage the mem - achieved. His enthusiasm for internationally oriented bers to make use of it and cooperate through EU funded research cooperation, his talent for recognising the research projects. potential use of mathematical tools, and his cooperative skills make him a young researcher of outstanding quality. For our leading role in Europe, lobbying is also important. I would like to have more contact with Brussels. We must act Tim has obtained some very impressive scientific results. His together, ERCIM, Informatics Europe, and, for instance, the PhD dissertation was awarded cum laude, which is conferred ACM Europe Council and EIT Digital. The cooperating to less than 5% of doctoral candidates at the Delft University organisations must become members of advisory bodies in of Technology. His dissertation won the 2014 Victor Lesser the European Commission, such as the CONNECT Advisory Distinguished Dissertation Runner-up Award in recognition Forum for Research and Innovation in ICT in Horizon 2020. of an exceptional and highly impressive dissertation in the I commit myself to recruiting members and to reinforce con - area of autonomous agents and multiagent systems. He was tacts with the EC. Together, as an association of organisa - shortlisted for the 2014 Artificial Intelligence Dissertation tions, we will endeavour to exert more influence on the poli - Award, which recognises the best doctoral dissertations in cies of the EU. For instance: to be heard at the creation of the the general area of artificial intelligence. His dissertation is European Innovation Council (EIC) and to be involved in the also published by the Springer Theses “the best of the best” preparations for FP9, the programme that follows H2020. We series, which recognises outstanding PhD research by need to be putting more effort into this area, and I intend to selecting the very best PhD theses from around the world for make this my personal mission. their scientific excellence and high impact on research.

In addition to lobbying, collaboration is also important. For Tim has made a real scientific impact with his work since instance, I believe that the annual Informatics Europe embarking on his PhD in 2010. He has already published meeting should continue to coincide with ERCIM meetings. over 40 articles, in collaboration with 35 researchers from 17 This requires much consultation and coordination. The other international institutes and five industrial partners. His semi-annual meeting of ERCIM should also be for research is published in the highest ranking publications in researchers, and not just the directors. his field, including the top journal Artificial Intelligence and top conferences such as IJCAI, ECAI, AAAI and AAMAS. One of ERCIM’s unique features is our combination of infor - His research on how artificial intelligence can help negotiate matics and mathematics, and these areas are becoming better deals for humans was recently featured in Science. His increasingly important for Europe. My personal opinion is achievements point to great promise for his future work and that Europe is not doing well: ICT companies are being sold for others who build on it. to the US and China, which means that we no longer have control over our data. I this is an unhealthy trend; citi - zens must have control over their own data and at least know who has them. We do not necessarily have to own our data but we must be able to control who has or uses them. Together, we need to work to benefit all European citizens. Together, we can conduct research that will bring us the innovations of the future – for the short and the long term. It is vital for the future of Europe that we are actively involved in helping to shape future innovations: that we do not lose the innovation capacity, as we are already losing ground to emerging economies and other world powers. China and the USA are out in front at the moment, and I think Europe must have a leading role, certainly in the field of long-term research for future innovations.

With ERCIM, I hope that we can contribute to this goal.

Jos Baeten, TimBarslaag(center)receivingtheawardfromtheERCIM General Director of CWI and President of ERCIM AISBL presidentDomenicoLaforenza(right)andERCIMpresident-elect JosBaeten.

ERCIM NEWS 112 January 2018 5 Joint ERCIM Actions

ERCIM Cor baayen • Frederik Diederichs (Fraunhofer • Giulio Rossetti (University of Pisa), IAO), nominated by Anette nominated by Fosca Giannotti Award 2017 Weisbecker Fraunhofer (IAO) (CNR) • Hadi Fanaee Tork (University of • Ville Salo (University of Turku), Winner: Oslo), nominated by Alípio Jorge nominated by Tuomo Tuikka (VTT) Tim Baarslag (CWI), nominated by (INESC) • Anna Ståhl, RI.SE SICS, nominated Eric Pauwels (CWI) • Eemil Lagerspetz (University of by Kristina Höök (RI.SE SICS) Helsinki), nominated by Tuomo • João Tiago Medeiros Paulo (INESC Honorary mention: Tuikka (VTT) TEC and University of Minho), Fabrice Ben Hamouda-Guichoux • Pierre Lairez (Inria), nominated by nominated byJosé Pereira (INESC (IBM Research), nominated by David Bertrand Braunschweig (Inria) TEC and University of Minho) Pointcheval (Inria) • Britta Meixner (CWI), nominated • Vassilis Vassiliades (Inria), by Pablo Cesar (CWI) nominated by Chris Christodoulou Finalists: • Iason Oikonomidis (ICS-FORTH), (University of Cyprus) • Markus Borg (RI.SE SICS), nominated by Antonis Argyros • Daniel Weber (Fraunhofer IGD), nominated by Jakob Axelsson (ICS-FORTH) nominated by Arjan Kuijper (RI.SE SICS) (Fraunhofer IGD).

Since 2016, Tim Baarslag has been working as a researcher Tim was the first to untangle the connection between negoti - at CWI, where he studies negotiation strategies for smart ation performance and opponent learning. In acknowledge - energy cooperatives. He collaborates closely with key smart ment of the originality of this work, he received the Best grid stakeholders to develop his negotiation results into user- Paper Award (out of 101 papers) at the 2013 adaptable energy trading technology. His negotiation algo - IEEE/WIC/ACM International Conference on Intelligent rithms support energy trading in a sustainable energy com - Agent Technology. Moreover, he was the first to combine a munity ‘Schoonschip’ in Amsterdam as part of the ERA-Net number of theoretical results in search theory, such as Grid-Friends project. He has recently received a personal optimal stopping and Pandora’s Problem, to formulate new ‘Veni’ grant for young, talented researchers from The and tested negotiation strategies. This resulted in the Best Netherlands Organisation for Scientific Research (NWO). Paper Award (out of 122 papers) for his results on optimal negotiation strategies during the 2015 IEEE/WIC/ACM As the lead developer of the negotiation environment International Conference on Intelligent Agent Technology. “Genius”, he contributed to an international platform for Moreover, a negotiating agent that successfully applied his research on automated negotiation agents, which is down - open source dissertation framework won The 2013 loaded more than 100 times a week and is used by more than International Automated Negotiating Agents Competition 20 research institutes all over the world, including Harvard (ANAC). University and Massachusetts Institute of Technology (MIT). Additionally, he is part of the organising team of the Tim is also actively engaged in the broader computing com - International Automated Negotiating Agent Competition munity and highly involved with the most eminent (ANAC), which has had more than 100 international partici - researchers within computer science and mathematics. In pants and is held in conjunction with the leading artificial 2016, he was selected as one of 200 highly talented, preemi - and multi-agent conferences in his field (AAMAS and nent young researchers in mathematics and computer scien - IJCAI). Not only have his results and the competition pro - tists from over 50 countries for the Heidelberg Laureate vided a state-of-the-art repository of automated negotiators, Forum scholarship. In 2017, he was selected from over 300 they also continue to steer the automated negotiation applicants as an outstanding early career individual in the research agenda. field of computing to take part in the ACM Future of Computing Academy and attend the 2017 Turing Award cer - Tim’s work provides a unique blend of mathematically emony, in an effort to support and foster the next generation optimal results and application-driven research with a strong of computing professionals. potential for knowledge utilisation. In collaboration with MIT CSAIL and the University of Southampton, he has See also Tim’s article “Computers that Negotiate on Our developed new mathematical models that enable essential Behalf” on page 34 in this issue. next steps for conducting privacy negotiations feasibly. Furthermore, together with multiple UK partners including Link: The Open University, Tim helps to design Internet of Things ERCIM Cor Baayen Award: solutions that enable negotiable data access for future data https://www.ercim.eu/human-capital/cor-baayen-award marketplaces. His research has influenced the negotiation architecture that will be used to assess and mitigate privacy trade-offs in existing Internet of Things applications and platforms.

6 ERCIM NEWS 112 January 2018 ERCIM Established Working Group on blockchain technology

ERCIM established a new Working Group Blockchain Technology to study the potential of this technology for a range of application fields in industry and public administra - tion. First chairperson of the Working Group is Wolfgang Prinz from the Fraunhofer Institute for Applied Information Technology FIT. ERCIM “Alain bensoussan” Following up on the ERCIM workshop on Blockchain fellowship Programme Technology in May 2017 in Paris, ERCIM now established the new Working Group Blockchain Technology. It has ERCIM offers fellowships for PhD holders from all over become evident that blockchain technology is being investi - the world. Topics cover most disciplines in Computer gated in various areas of computer science research. That Science, Information Technology, and Applied includes peer-to-peer networks, distributed systems, cryp - Mathematics. Fellowships are of 12 months duration, tography, algorithms for consensus building and validation spent in one ERCIM member institute. Fellowships are as well as for modeling processes and business models. In proposed according to the needs of the member institutes addition to these basic technologies, the Working Group will and the available funding. also study applications in relevant fields like Internet-of- Things, supply chains, energy and Smart Grid, the media, the Application deadlines for the next rounds: 30 April medical field and the financial industry. and 30 September 2018 “We want to establish blockchain technology as a new field More information: http://fellowship.ercim.eu/ of computer science research in Europe. Our working group will strengthen the young community of European blockchain researchers and foster interdisciplinary coopera - HoRIZoN 2020 tion and exchange”, Wolfgang Prinz stated. The Working Group will also establish a roadmap docu - Project Management menting ongoing research and open research questions in the rapidly moving field of blockchain technology. Another ele - A European project can be a richly rewarding tool for ment of its mission is to build a blockchain research commu - pushing your research or innovation activities to the state-of- nity that will be able to successfully initiate and carry out the-art and beyond. Through ERCIM, our member institutes collaborative research projects on a European level. have participated in more than 80 projects funded by the European Commission in the ICT domain, by carrying out For its first year the group is planning several meetings and joint research activities while the ERCIM Office success - editorial projects in accordance with the objectives of the fully manages the complexity of the project administration, group. The next major event will be a workshop organized by finances and outreach. members of the group, to be held May 8-9, 2018 in Amsterdam, in conjunction with the ERCIM spring meet - The ERCIM Office has recognized expertise in a full range ings. of services, including identification of funding opportunities, recruitment of project partners, proposal writing and project Researchers from ERCIM member institutions and other negotiation, contractual and consortium management, com - organizations are cordially invited to join the new ERCIM munications and systems support, organization of attractive working group Blockchain Technology. events, from team meetings to large-scale workshops and conferences, support for the dissemination of results. Link: ERCIM Blockchain WG: How does it work in practice? https://wiki.ercim.eu/wg/BlockchainTechnology/ Contact the ERCIM Office to present your project idea and a panel of experts will review your idea and provide recom - Please contact: mendations. If the ERCIM Office expresses its interest to For additional information please contact the workgroup participate, it will assist the project consortium as described chair Wolfgang Prinz above, either as project coordinator or project partner. Fraunhofer FIT, Germany [email protected] Please contact: Philippe Rohou, ERCIM Project Group Manager [email protected]

ERCIM NEWS 112 January 2018 7 Special theme: Quantum Computing

Introduction to the Special Theme Quantum Computation and Information

by Jop Briët (CWI) and Simon Perdrix (CNRS, LORIA)

For more than a century now, we’ve understood that we live in a quantum world. Even though quantum mechanics cannot be ignored during the development of atomic scale components of everyday computers, the computations they perform are governed, like the Turing machine, by the laws of classical Newtonian mechanics. But the most striking and exotic features of quantum mechanics, superposition and entanglement, currently play no part in every-day information processing. This is about to change – and in some specialised applications, already has. In academia, the field of quantum computation has been growing explosively since its inception in the 1980s and the importance of these devices is widely recognised by industry and governments. Big players in the tech industry like IBM and Google frequently announce that they have built yet a larger rudimentary quantum computation device and in 2016 the European Commission launched a one-billion Euro Flagship Initiative on Quantum Technologies.

The development of this technology has computers could do. Think for Basic features of quantum both a hardware and a software compo - instance about machine learning, computation nent, both of which have been the sub - search and optimisation and tackling and superpositions jects of intense research for the past few number-theoretic problems (relevant The basic building block for classical decades. On the hardware side, past to cryptography). computation is the bit, a physical efforts went into controlling and storing • Cryptography , important because system that can be toggled between one small numbers (5-10) of qubits, the fun - larger scale quantum computers will of two possible states, 0 and 1. For damental building blocks for quantum break our current cryptosystems, but quantum computers, this building block computation. The current state of the art also hold the key for future cryp - is fundamentally different. It is the is that this can be done for arrays of tosystems. , short for quantum bit. A qubit is a about fifty qubits. Although a seemingly • Quantum software framework, physical system that can be in a sort of modest scale, this is where things including intermediate state given by one of the become very interesting, because languages, together with formal and infinitely many possible superpositions quantum systems of 50-100 qubits logical methods to ease the use of of two basis states. If the two basis cannot be simulated on our current clas - quantum computers and understand states are represented by two-dimen - sical computers and so hold the possi - the fundamental structures of quan - sional row vectors (1,0) and (0,1), then bility of harnessing a computational tum information processing. a superposition is a complex Euclidean power we have not seen before. A more • science in gen - unit vector of the form ( x,y ) where x and long-term goal (in the order of a few eral, which is to provide the mathe - y are complex numbers, referred to as decades) is to scale-up further to mil - matical theory behind quantum probability amplitudes, whose absolute lions of qubits. This scale is what the information processing and is impor - values squared sum to 1. A measure - NSA is worried about, because quantum tant for the development of error cor - ment of the qubit results in finding it in computers of this size could break most rection schemes, understanding the first or second state with probability modern-day cryptography. counterintuitive quantum phenome - |x|2 or | y|2, respectively. Toggling of na and the physical theory behind the quantum states can be done via linear, Hardware, however, is not much good hardware. length-preserving operations, in other without interesting software. Some words, unitary transformations. A meas - important research areas on this side of The articles in this special theme offer a urement performed after a unitary oper - the development include: more detailed look into the complexi - ation is also referred to as doing a meas - • Quantum simulation , which includes ties of some of the above-mentioned urement in a different basis. Whereas research on applications of medium- aspects of the emerging paradigm shift the state of n classical bits can be repre - sized quantum computers. Potential currently happening in computation. sented by an n-bit string, the state of an application areas include chemistry Below you will find lightning-fast n-qubit system is in general given by a and materials science. introductions to the basic concepts superposition of 2 n basis states: a com - • Algorithms and complexity, which appearing in the articles. plex 2 n-dimensional Euclidean unit explores what larger-scale quantum vector. The observation that n-qubit

8 ERCIM NEWS 112 January 2018 states require an exponential number of amplitudes describing the overall state tion protocol, an unconditionally secure parameters to be described is what of the register. These cancellations can communication protocol relying on the makes simulating quantum-mechanical in turn cause large superpositions to laws of quantum mechanics. In this pro - systems hard to do on classical com - quickly converge into a state that, when tocol, Alice wants to a random puters. It is also one of the key features measured, with near-certainty gives key with Bob, to do so she sends of quantum mechanics that gives only a single possible outcome. Two random bits encoded into randomly quantum computers their power. early, but still important, examples chosen basis among two complemen - demonstrating the power of this phe - tary basis. Complementary basis are Entanglement nomenon are Shor’s factoring algorithm subject to the uncertainty principle: if a Arguably the most striking features of and Grover’s search algorithm. Shor’s bit of information is encoded into one of quantum mechanics is entanglement, algorithm factors a given number into the basis, measuring according to the which manifests itself when two or its prime-number components exponen - other one produces an random bit, more quantum systems are measured tially faster than the best-known clas - uncorrelated with the encoded informa - locally in one of two or more possible sical algorithm to date, with dramatic tion. Roughly speaking a spy who wants bases. As an argument against quantum, consequences for important crypto - to observe the sent qubits will not only it was observed by Einstein, Podolsky graphic schemes (see below). Grover’s get no information if he does not guess and Rosen that compound systems algorithm finds the location of a 1 with the appropriate basis, but will be allow superpositions that can result in good probability in a given n-bit detected by Alice and Bob with high measurement statistics that defy clas - sequence, provided there is one, using a probability. The protocol is relatively sical explanation. Celebrated work of number of basic computational steps simple to implement, and has been com - Bell later showed that one can actually given by roughly the square root of n. mercialised. Notice however, the proof test this feature experimentally using The best-possible classical algorithm that the protocol is actually uncondi - what is nowadays referred as a Bell test. needs roughly n steps in the worst case, tionally secure took 15 years! The most basic example of such a test however. can be cast as a game involving two Scalability and error correction players, Alice and Bob, and a referee. Quantum cryptography One of the major challenges in the quest The referee picks two bits at random Shor’s algorithm can break the most for a quantum computer is its scala - and sends these to Alice and Bob, important cryptographic schemes we bility. Prototypes of quantum computers respectively. Without communicating, have today, which are based on the already exist but their memory, subject and thus without knowing the other’s assumption that factoring is hard. A to decoherence, is limit to a few dozen bit value, the players each return a fully operational large-scale quantum qubits. The scalability of quantum com - binary answer to the referee. They win computer shatters this assumption. This puters is not only a technological chal - the game if the sum of the answers means that alternative cryptography is lenge, error correcting code are crucial: modulo 2 equals the product of the needed immediately, as some of today’s a large scale robust quantum computer questions. A simple calculation shows information may need to be kept secret requires that the quality of the quantum that if they players are constrained by even after quantum computers are built. device should meet the maximal the laws of classical physics, then they Multiple lines of research on post- amount of errors quantum codes can can win this game with probability at quantum cryptography address this correct. most 3/4. However, by performing local issue. On the one hand, one can try to measurements on a two-qubit “EPR look for problems that might even be Please contact: pair”, the players can produce a proba - hard for quantum computers, an impor - Jop Briët bility distribution over their answer tant motivation for lattice-based cryp - CWI, The Netherlands pairs with which they win with proba - tography. On the other hand, quantum [email protected] bility roughly 0,85! mechanics itself offers alternatives too, as was pointed out already long before Simon Perdix Quantum algorithms Shor’s discovery. Wiesner circa 1970 LORIA, CNRS, Inria Mocqua, Quantum algorithms consist of a care - introduced a quantum money scheme, a Université de Lorraine, France fully chosen sequence of unitary trans - proposal where bank notes are unfalsifi - [email protected] formations that are applied one by one able thanks to the presence of qubits on to a register of qubits, followed in the it. This was the first attempt to use end by a measurement to determine an quantum mechanics in a cryptographic output. A crucial property of unitary protocol. More than a decade later, in transformations is that they can cause 1984, Bennett and Brassard introduced cancellations among the probability the revolutionary quantum key distribu -

ERCIM NEWS 112 January 2018 9 Special theme: Quantum Computing

How Classical beings Can test Quantum devices

by Stacey Jeffery (CWI)

As the race to build the first quantum computer heats up, we can soon expect some lab to claim to have a quantum computer. How will they prove that what they have built is truly a quantum computer?

It is already possible to buy quantum the physical systems being tested. If the However, there are means of verifying hardware for certain cryptographic provers are “honest” and follow the pro - certain aspects of quantum mechanics tasks, such as random number genera - tocol – that is, they behave as predicted that do not require the experimenter to tion and key distribution. In crypto - – the verifier should learn the result of have a trusted quantum device, called graphic scenarios, being able to test that the quantum computation. However, if Bell tests . In a Bell test, two provers play your devices behave as advertised is the provers are deviating from the pro - a game with a verifier. The verifier asks clearly of paramount importance. tocol, the verifier should detect this and each prover a question, and they must “reject”. each return an answer, without commu - The first commercial quantum com - nicating during the game. What makes a puters are likely to run as shared servers, Verifying quantum computations is Bell test special is that they can win the where clients can pay to have a quantum related to the more fundamental task of game with higher probability if they computation run. This is another sce - experimentally verifying quantum share a quantum resource called entan - nario where we would like the client, mechanics. A quantum system has an glement than if they are classical. Thus, who does not have a quantum computer, internal state that an observer cannot such a game offers a way to experimen - to have some guarantee that the correct perceive directly. To learn about this tally test quantum mechanics. computation has been performed. state, a quantum measurement can be performed, giving some incomplete Analogous to the situation in testing These scenarios are captured by a verifi - information. If an experimentalist quantum mechanics, testing a quantum cation protocol , in which a verifier , who hypothesises that a particular quantum computer can be accomplished in either would like to run some quantum compu - system has a particular state, this can be of two regimes. In the quantum-verifier tation, interacts with one or more verified by performing measurements, regime , the verifier must have a simple provers who have a quantum computer. but this presupposes some trusted trusted quantum device, like a measure - By the end of the protocol, the verifier quantum measurement device. If one ment device. In this regime, several effi - either “accepts” or “rejects” the output wants to verify the theory of quantum cient verification protocols are known. of the computation. The verifier might mechanics, one cannot circularly However, the requirement that the veri - represent an experimenter testing a assume that the measurement device fier have a trusted quantum device is a quantum system, and the provers a per - behaves as predicted by the theory big drawback. sonification of nature, or more precisely, quantum mechanics. In the two-prover regime , analogous to a Bell test, a classical verifier interacts with two provers who do not communi - cate. The first protocol in this regime was a significant breakthrough, pro - viding, for the first time, a method for a

Verifier Prover classical verifier to verify any quantum

Figure1:Quantum-verifierRegime.Averifierwithasimplequantum deviceinteractswithaproverwithafullquantumcomputer.

Prover V Prover P

Verifier Figure3:Ournewtwo-proverprotocol.Theproversplaytheroleofthe Figure2:Two-proverRegime.Aclassicalverifierinteractswithtwo verifierandproverfromBroadbent’sProtocol,whichisaquantum- quantumprovers. verifierprotocol.

10 ERCIM NEWS 112 January 2018 computation [3]. However, the protocol tively. We call our provers Prover V, for Links: had the major disadvantage of requiring verifier, and Prover P, for prover. CWI’s algorithms and complexity resources that scale like g8192 to verifi - group: https://kwz.me/hBD ably implement a of By the properties of Broadbent’s QuSoft: http://www.qusoft.org/ size g. For comparison, there are an esti - Protocol, we can use Prover V to test mated 2 286 elementary particles in the Prover P, to make sure he is following References: universe. Subsequent improvements the protocol. Then it only remains to [1] A. Broadbent: “How to Verify a decreased this overhead to g2048 , but ensure that Prover V is, in fact, fol - Quantum Computation” Arxiv preprint this is still thoroughly impractical, even lowing the protocol. We develop new arXiv:1509.09180, 2015. for a quantum circuit of size g = 2. Bell tests that are used to make Prover P [2] A. Coladangelo, A. Grilo, S. test Prover V. While the classical verifier Jeffery, T. Vidick: “Verifier-on-a- In collaboration with researchers from may not be powerful enough to keep a Leash: new schemes for verifiable Université Paris Diderot and Caltech, quantum prover in check, she can use delegated quantum computation, with we presented the first efficient protocol the two provers to control one another. quasilinear resources” Arxiv preprint in the two-prover regime [2]. This pro - arXiv:1708.07359, 2017. tocol requires resources that scale like In the future, we hope to see experi - [3] B. W. Reichardt, F. Unger and U. O(g log g) to verify a quantum circuit of mental realisations of our protocol, Vazirani: “Classical command of size g. which is possible for the first time, due quantum systems” Nature 496, 456- to its efficiency. Its near-optimal effi - 460, 2013. We begin with a quantum-verifier pro - ciency means that verifying a particular tocol due to Broadbent [1]. The provers quantum computation does not require Please contact: in our protocol are asked to simulate much more resources than simply Stacey Jeffery, CWI and QuSoft, Broadbent's verifier and prover, respec - implementing that computation. [email protected]

Keeping Quantum Computers Honest (or verification of Quantum Computing) by Alexandru Gheorghiu (University of Edinburgh) and Elham Kashefi (University of Edinburgh, CNRS)

Quantum computers promise to efficiently solve not only problems believed to be intractable for classical computers, but also problems for which verifying the solution is also intractable. How then, can one check whether quantum computers are indeed producing correct results? We propose a protocol to answer this question.

Quantum information theory has radi - numbers which is exponentially faster involved, and because predicting the cally altered our perspective about than the best known classical factoring outcome of the experiment requires quantum mechanics. Initially, research algorithms. exponential resources, it quickly into quantum mechanics was devoted to becomes infeasible to calculate the pos - explaining phenomena as they are But having a device which can solve sible results of the experiment without observed in nature. But the focus then problems exponentially faster than clas - resorting to lax approximations. changed to designing and creating sical computers raises an interesting Verification of quantum computation quantum systems for computation, question: can a classical computer effi - would therefore allow for a new test of information processing, communica - ciently verify the results produced by quantum mechanics, a test in the regime tion, and cryptography among many this device? At first, one might be of high complexity. other tasks. In particular, what became tempted to dismiss this question and say clear was that quantum interference - that as long as each component of a There is another important reason for “the heart of quantum mechanics”, as quantum computer has been tested and verifying quantum computations, Richard Feynman described it - can be works correctly, there is no need to having to do with cryptography. The harnessed for quantum computation. worry about the validity of the device's first quantum computers are likely to be Algorithms running on a hypothetical results. However, the point of verifica - servers, to which clients can connect quantum computer would be able to tion is much more profound. Quantum through the Internet. We can already see solve problems by creating an interfer - computers would provide one of the an instance of this with the recent 5- ence pattern of different computational most stringent tests of the laws of qubit and 16-qubit devices that IBM has branches. This can lead to an exponen - quantum mechanics. While numerous made available to the general public tial saving in the amount of resources experiments involving quantum sys - [L1]. When larger devices become used by a , when tems have already been performed to a available, users will wish to delegate compared to the best known classical remarkable precision, they all utilized complex computations to them. algorithms. The most famous example relatively few degrees of freedom. But However, in such a distributed environ - of this is Shor's algorithm for factoring when many degrees of freedom are ment, malicious agents might perform

ERCIM NEWS 112 January 2018 11 Special theme: Quantum Computing

Figure1:Device-independentverificationprotocol.The client,orverifier,willinstructboththemeasurementdevice andtheservertomeasureentangledqubits.Thestatisticsof thesemeasurementsarethencheckedbytheverifier.All communicationwiththequantumdevicesisclassical.

man-in-the-middle attacks or compro - Vazirani, known as rigidity of non-local imperfections, as well as other issues. mise the remote server. The clients correlations [4]. Non-local correlations Addressing all these questions is a key would then need a means to check the are correlations between responses of challenge for both theorists and experi - validity of the server's responses. In non-communicating parties that cannot mentalists and their resolution will fact, in this setting, users might also be reproduced classically, unless the shape the landscape of the emerging wish to keep their data hidden even parties are allowed to communicate. Quantum Internet. from the quantum computer itself, as it Such correlations can be produced, might involve sensitive or classified quantum mechanically, through a suit - Link: information. able strategy for measuring certain [L1] https://kwz.me/hBv entangled states. The rigidity result is So can one verify quantum computa - essentially a converse to this. It states References: tions while also maintaining the secrecy that certain non-local correlations can [1] A. Broadbent, J.F. Fitzsimons, E. of the client's input? The answer is yes. only be produced by a particular, unique Kashefi: “Universal blind quantum In fact, the client's ability to keep the strategy. Observing such correlations computation”, in Proc. of FOCS input hidden is what makes verification between non-communicating devices ‘09, IEEE Computer Society possible. This was shown by Fitzsimons then implies that the devices are (2009) 517 – 526. and Kashefi when they proposed a veri - behaving according to this fixed [2] J.F. Fitzsimons, E. Kashefi: fication protocol based on a crypto - strategy. What is remarkable about this “Unconditionally verifiable blind graphic primitive known as Universal result is that it only requires examining quantum computation”, Phys. Rev. Blind Quantum Computation (UBQC) the outputs of the devices, without A 96 (2017) 012303. [1,2]. In UBQC, a client that can pre - assuming anything about their inner [3] A. Gheorghiu, E. Kashefi, P. pare single qubits has the ability to dele - workings. Wallden: “Robustness and device gate quantum computations to a server, The protocol then works as follows: the independence of verifiable blind in such a way that the server is oblivious client has an untrusted device for meas - quantum computing”, New Journal to the computation being performed. To uring single qubits and is also commu - of Physics 17(8) (2015) 083040. do verification, the client can then nicating classically with the quantum [4] B.W. Reichardt, F. Unger, U. exploit this property by embedding tests server. By examining the outputs of the Vazirani: Classical command of in the computation, referred to as traps, two devices, it follows from the rigidity quantum systems. Nature which will fail if the server doesn't per - result that the client can check whether 496(7446) (2013) 456. form the correct computation. Of the two devices are sharing entangle - course, the problem with this approach ment and performing measurements as Please contact: is that the client needs to trust that the instructed. If so, the client leverages this Elham Kashefi, University of qubit preparation device works cor - and uses the entanglement to remotely Edinburgh, UK and CNRS, France rectly and produces the specified states. prepare single qubit states on the [email protected] But if, prior to the start of the protocol, a server's side. Finally, the client uses the malicious agent corrupts the prepara - trap-based scheme of Fitzsimons and Alexandru Gheorghiu tion device, the client could later be Kashefi to delegate and verify an arbi - University of Edinburgh, UK tricked into accepting incorrect results. trary quantum computation to the [email protected] server. To address this issue, we, together with Dr. Petros Wallden, at the University of Verification is an important milestone Edinburgh, proposed a verification pro - on the road to scalable quantum com - tocol which is device-independent [3]. puting technology. As we have seen, In other words, the client need not trust verification protocols exist even for the any of the quantum devices in the pro - most paranoid users. But even so, ques - tocol. This is achieved by using a pow - tions still remain regarding their opti - erful result of Reichardt, Unger and mality, their ability to tolerate noise and

12 ERCIM NEWS 112 January 2018 Experimental Requirements for Quantum Computational Supremacy by boson Sampling by Alex Neville and Chris Sparrow (University of Bristol)

Boson sampling has emerged as a leading candidate for demonstrating “quantum computational supremacy”. We have devised improved classic al algorithms to solve the problem, and shown that photon loss is likely to prevent a near-term demonstration of quantum computational supremacy by boson sampling.

The prospect of harnessing precisely error correction. Physically, the com - for up to 100 photons if we ran MIS on a engineered quantum systems to run plexity of the problem stems from the powerful supercomputer. Alongside algorithms which vastly outperform complex quantum interference of indis - this, we computed the expected time to their best classical counterparts has gen - tinguishable photons; if we make the experimentally perform boson sampling erated a large amount of interest and photons distinguishable (say, by using as a function of the probability of a investment in the field of quantum com - photons of different colours) then the single photon surviving the experiment puting. However, a universal and fault- problem is no longer hard. (i.e., not getting lost from source to tolerant quantum computer is unlikely to detector). We found that it would be built soon. Considering this poten - Our best new classical boson sampling require more than 50 photons before the tially long wait, recently several new algorithm [3] is based on Metropolised classical runtime would exceed a week. problems have been proposed with the Independence Sampling (MIS), a To achieve this experimentally would purpose of demonstrating “quantum Markov chain Monte Carlo procedure not only require the ability to produce a computational supremacy”, which refers similar to the famous Metropolis- state of 50 indistinguishable photons at to the point at which a Hastings algorithm. We start a list by a high rate (the current record-holding performs a computational task that is proposing a sample value from some boson sampling experiment is with five beyond the capability of any classical easy-to-sample distribution, and con - photons), but also that all photons sur - computer [1], with specialised near-term tinue by iteratively proposing new vive the experiment with a probability quantum devices. values and either adding this new value greater than 50%. Considering that each to the list or repeating our previous photon would require to be injected in One such proposal is the boson sampling value. Which of these possibilities actu - to the circuit, pass through over 1000 problem introduced by Aaronson and ally occurs is determined by a carefully beamsplitters and be coupled out of the Arkhipov [2]. The problem consists of crafted acceptance rule which depends circuit to single photon detectors, this sampling from the output distribution of on the probabilities of the proposed and would amount to a significant engi - detection events generated when many current value occurring in both the neering breakthrough which is unlikely single photons are concurrently injected target distribution and the proposal dis - to be realised in the near future. in to a randomly chosen network of tribution, and guarantees that the list Our results not only provide a bench - linear optical components. A sample tends towards a genuine boson sam - mark for quantum supremacy by boson value is generated by recording where pling sample. sampling, but also highlight how signif - photons were detected at the end of the icant the experimental improvement linear optical network. The probability We identified the corresponding distri - must be in order to achieve this. for each possible output in the experi - bution for distinguishable particles at ment is related to a matrix function the output of the linear optical network Link: known as the permanent, which is in as a proposal distribution which pro - https://kwz.me/hB1 general especially hard to compute. By vides rapid convergence to the boson making the plausible conjecture that the sampling distribution. Although the References: permanent is hard to approximate for probabilities in this distribution are also [1] A. Harrow, A Montanaro: “Quantum Gaussian random matrices, as well as given by matrix permanents (albeit dif - computational supremacy”, Nature another reasonable conjecture about the ferent matrices), the distribution can be 549 (2017), 203–209. distribution of these permanents, efficiently sampled. Using this, we [2] S. Aaronson, A. Arkhipov: “The com - Aaronson and Arkhipov were able to found strong numerical evidence that putational complexity of linear optics”, show that an efficient classical algo - computing just 200 matrix permanents Theory Comput. 9, 143–252, 2013. rithm for even approximate Boson sam - is enough to generate a boson sampling [3] A. Neville, C. Sparrow, et al.: pling would lead to the collapse of the value via MIS for a problem size of up “Classical boson sampling algo - polynomial hierarchy of complexity to 30 photons, roughly amounting to a rithms with superior performance to classes to its third level – something speed-up of 50 orders of magnitude near-term experiments”, Nature considered extremely unlikely in com - over the best previously suggested clas - Physics 13, 1153–1157 (2017). putational complexity theory. The result sical algorithm at this problem size. applying to approximate boson sam - By assuming that 200 matrix permanent Please contact: pling is key, as it allows for the possi - computations suffice to produce a good Alex Neville, Chris Sparrow bility of solving the boson sampling sample for larger photon numbers, we University of Bristol, UK problem with realistic quantum experi - were able to predict the amount of time [email protected], ments, without the need for quantum it would take to solve boson sampling [email protected]

ERCIM NEWS 112 January 2018 13 Special theme: Quantum Computing

from Classical to Quantum Information – or: When You Have Less than No Uncertainty

by Serge Fehr (CWI)

Over the last few years, significant progress has been made in understanding the peculiar behaviour of quantum information. An important step in this direction was taken with the discovery of the quantum Rényi entropy. This understanding will be vital in a possible future quantum information society, where quantum techniques are used to store, communicate, process and protect information.

Information theory is the area of com - puter science that develops and studies abstract mathematical measures of “information” – or, from a more pes - simistic perspective, measures of “uncertainty”, which capture the lack of information. It is clear that when you toss a coin there will be uncertainty in the outcome: it can be either “head” or “tail”, and you have no clue what it will be. Similarly, there is uncertainty in the face that will show up when you throw a dice. It is even intuitively clear that there is more uncertainty in the latter than in the coin toss. However, such comparisons become less clear for more today’s information and communication extension of its classical commutative complicated cases. If we want to com - infrastructure. counterpart. This insight can serve as a pare, say, tossing three coins on the one guideline for coming up with quantum hand with throwing two dice and taking What makes information theory very versions of classical information meas - the sum of the two faces on the other powerful is its independence of how ures, but it also shows a typical predica - hand, it is not immediately clear in information is physically represented: ment: a commutative expression can be which of the two there is more uncer - whether the information is represented generalised in various ways into a non- tainty: there are fewer possible out - by coins that show “head” or “tail”, or commutative one. For instance, an comes in the former, namely eight, but, by the tiny indentations on a DVD, or expression like A⁵B⁴ can be generalised on the other hand, the eleven possible whether the information is stored on a to A⁵B⁴ or to B⁴A⁵ for non-commuting outcomes in the latter are biased. flash drive or communicated over A and B, or to ABABABABA, or to WiFi, the predictions of information B²A⁵B², etc. One of the challenging Information theory offers quantitative theory hold universally – well, until we questions is to understand which of the measures that express precisely how hit the realm of quantum mechanics. If, possible generalisations of classical much uncertainty there is. Similarly, it say, we encode information into the information measures are suitable also offers measures of conditional polarisation of photos, then informa - measures of quantum information and uncertainty given that one holds some tion starts to behave very differently. have operational significance. “side information”. For instance, how Therefore, a quantum version of infor - much uncertainty is there in the faces of mation theory is necessary in order to Building upon new insights and new the two dice given that I know the sum rigorously study the behaviour of infor - results [1] that we discovered on the of the two? How much uncertainty is mation in the quantum realm, and, for classical notion of Rényi entropy, and in there in a message that was communi - instance, to be able to quantify the collaboration with several partners, we cated over a noisy channel given that I amount of information an eavesdropper succeeded in lifting the entire family of hold the received noisy version? How may have on secret data when the data Rényi entropies to the quantum setting much uncertainty is there in a digital is protected by means of quantum cryp - [2,3]. The Rényi entropies form a con - photo given that I hold a compressed tography, or to quantify the amount of tinuous spectrum of information meas - version? How much uncertainty does an error correction needed in order to ures and cover many important special eavesdropper have on secret data given counter the loss of information in cases; as such, our extension to the that he got to see an encryption? quantum computation caused by deco - quantum setting offers a whole range of Information theory allows us to answer herence. new quantum information measures. such questions in a precise manner and We showed that our newly proposed to make rigorous predictions about the From a mathematical perspective, given definition satisfies various mathemat - behaviour of information in all kinds of that quantum mechanics is described by ical properties that one would expect information processing tasks. As such, non-commuting mathematical objects, from a good notion of information and information theory had – and still has – a quantum information theory can be which make it convenient to work with major impact on the development of understood as a non-commutative the definition. It is due to these that our

14 ERCIM NEWS 112 January 2018 quantum Rényi entropies have quickly under different measurements. References: turned into an indispensable tool for However, if you are given another [1] S. Fehr, S. Berens: “On the studying the behaviour of quantum quantum state that is entangled with the Conditional Rényi Entropy”, in information in various contexts. state of interest, then you can actually IEEE Trans. Inf. Theory predict the behaviour of the state of 60(11):6801 (2014). One very odd aspect of quantum infor - interest under any measurement by [2] M. Müller-Lennert, F. Dupuis, O. mation is that uncertainty may become means of performing the same meas - Szehr, S. Fehr, M. Tomamichel: negative: it may be that you have less urement on your state. Indeed, by what “On Quantum Rényi Entropies: A than no uncertainty in your target of Einstein referred to as “spooky action New Generalization and Some interest. This peculiarity is an artifact at a distance”, the measurement on Properties”, in J. Math. Phys of entanglement, which is one of the your entangled state will instanta - 54:122203 (2013). most bizarre features of quantum neously affect the other state as to pro - [3] M. Wilde, A. Winter, D. Yang: mechanics. Entanglement is a form of duce the same measurement outcome. “Strong Converse for the Classical correlation between quantum informa - Capacity of Entanglement- tion that has no classical counterpart. A The above aspect nicely illustrates that Breaking and Hadamard Channels sensible explanation of negative uncer - quantum information theory is much via a Sandwiched Rényi Relative tainty can be given as follows. By more than a means for understanding Entropy”, in Commun. Math. Phys Heisenberg’s uncertainty principle, the behaviour of information within 331:593 (2014). even when given full information in the possible future quantum communica - form of a perfect description of the tion and computation devices: it sheds Please contact: quantum state of interest, there is still light on the very foundations of Serge Fehr, CWI, The Netherlands uncertainty in how the state behaves quantum mechanics itself. +31 20 592 42 57, [email protected]

Preparing ourselves for the threats of the Post-Quantum Era by Thijs Veugen (TNO and CWI), Thomas Attema (TNO), Maran van Heesch (TNO), and Léo Ducas (CWI)

In the post-quantum era, most of the currently used cryptography is no longer secure due to quantum attacks. Cryptographers are working on several new branches of cryptography that are expected to remain secure in the presence of a universal quantum computer. Lattice-based cryptography is currently the most promising of these branches. The new European PROMETHEUS project will develop the most secure design and implementations of lattice-based cryptographic systems. Exploitation of the project results will be stimulated by demonstrating and validating the techniques in industry-relevant environments.

Most current day cryptosystems are To ensure secure communication in the quantum-safe cryptographic systems. based on two computationally hard presence of a universal quantum com - The twelve project partners (ENS Lyon, problems: factoring integers and com - puter, for many years cryptographers ORANGE SA, CWI Amsterdam, IBM puting the discrete logarithms in finite have been working on constructing Research, RHU London, RU Bochem, groups. Advances in solving these prob - cryptographic schemes based on other Scytl Barcelona, Thales, TNO, UPC lems (classically) and increased clas - computational problems since many Barcelona, University Rennes, WIS sical computational power form a threat years. Their efforts have led to the fol - Rehovot) will focus their efforts on lat - that is relatively easy to diminish by lowing six possible building blocks for tice-based cryptography. A fundamental ‘simply’ increasing the key sizes. A cryptographic schemes: hash trees, property of lattice-based cryptographic more serious threat was revealed in 1994 error-correcting codes, lattices, multi - schemes is that their security can be already, when Shor [1] had proven that variate equations, supersingular elliptic reduced to well-studied computational quantum algorithms are able to effi - curve isogenies, and even quantum problems, which is not necessarily the ciently solve both these problems. While physics. These cryptographic building case for other post-quantum mecha - the requirement for a truly universal blocks fall within the realm of so-called nisms. quantum computer to execute these “post-quantum cryptography”, the attacks has long kept us secure, recent study of cryptographic algorithms that An n-dimensional lattice L is a set of advances in quantum computing once are secure against attacks by a quantum integer linear combinations of n inde - again highlight our dependency on these computer. pendent vectors. The grid displayed in computational problems. In short, the Figure 1 represents a lattice L. An arrival of a quantum computer will leave PROMETHEUS is a new, four-year example of a computationally hard lat - commonly used cryptosystems such as European H2020 project (starting in tice problem is the closest vector RSA, DH and ECDH insecure. January 2018) aiming to provide a secure problem (CVP): given a lattice L and a design and implementation of new and random point y (not necessarily an ele -

ERCIM NEWS 112 January 2018 15 Special theme: Quantum Computing

Figure1:Illustrationof atwo-dimensionallattice.

ment of the lattice), find the lattice ele - much more than protecting private ment x closest to y. For higher dimen - information over insecure communica - Links: sional lattices this problem soon tion channels. With its digital signa - [L1] https://kwz.me/hB2 becomes very hard to solve. In fact, no tures, commitment schemes, and homo - [L2] https://kwz.me/hB3 (quantum) algorithms have been found morphic properties, lattice-based cryp - [L3] https://kwz.me/hB5 yet solving this problem efficiently. tography offers a wide variety of appli - cations. To enhance the applicability References: Research in using this and other hard and the adaptation of these techniques, [1] P.W. Shor: “Algorithms for lattice problems for cryptographic pur - four use-cases will be studied. By quantum computation: Discrete poses began with the publication of the means of these use-cases logarithms and factoring”, in Proc. public key encryption scheme by Ajtai PROMETHEUS aims to cover the of SFCS ‘94, 1994. and Dwork in 1997 [2]. Improvements entire range from theory to application. [2] M. Ajtai, C. Dwork: “A public-key to this scheme and new schemes fol - cryptosystem with worst- lowed, trying to make use of additional The first use case will focus on con - case/average-case equivalence”, structures in specific types of lattices. structing an anonymous credential Proc. of the 29th annual ACM The first lattice-based cryptographic system, which allows a user to prove to symposium on Theory of schemes for public-key encryption, sig - a service provider that he owns a certain Computing. ACM, 1997. natures and key-exchange have been attribute (e.g., driving licence), while [3] E. Alkim, L. Ducas, T. proposed, and are considered for stan - minimising the information given to Pöppelmann, P. Schwabe: “Post- dardization [L1]. A recently proposed third parties, then protecting the user’s quantum Key Exchange-A New key-exchange protocol [3] (partly privacy. In the second use case tech - Hope”, USENIX Security developed at CWI) has successfully nology will be developed that allows Symposium, 2016. been implemented and tested by Google users to make secure, privacy-friendly [4] R. Cramer, L. Ducas, B. in the Chrome web browser [L2], and contactless transactions, for a long-term Wesolowski: ‘’Short Stickelberger was awarded the Facebook Internet use. In the third use case, a long-term Class Relations and application to Defense prize [L3]. secure e-voting system will be devel - Ideal-SVP’’, IACR, Eurocrypt oped. Lastly, in the fourth use case, 2017. One particular challenge for the transfer PROMETHEUS will develop quantum- from theory to practice lies in the choice safe homomorphic encryption tech - Please contact: of parameters for those schemes: while niques, and use them to develop a Thijs Veugen, TNO, The Netherlands lattice problems become ‘hard’ with secure cyber threat-intelligence sharing +31888667314, [email protected] larger dimensions, predicting precisely mechanism. how hard they are (e.g., it will take 100 years to solve with a cluster of 10,000 All four use-cases aim for long-term GPUs) remains difficult and uncertain. security in the quantum era, requiring The issue becomes even more delicate the cryptographic building blocks to be when considering quantum algorithms, quantum safe. Many of these techniques especially in the light of recent quantum already exist in a quantum vulnerable algorithms specialised for “ideal lat - setting. Introducing long-term security tices” [4]. This is the main issue for by mitigating quantum threats is the which CWI’s crypto group will provide core of our innovation. By covering the its expertise. entire range from theory to application and building demonstrators, the The PROMETHEUS project recognises exploitation of the project results will that modern day cryptography entails be stimulated.

16 ERCIM NEWS 112 January 2018 Quantum Cryptography beyond Key distribution by Georgios M. Nikolopoulos (IESL-FORTH, Greece)

Quantum cryptography is the science of exploiting fundamental effects and principles of quantum physics, in the development of cryptographic protocols that are secure against the most malicious adversaries allowed by the laws of physics, the “quantum adversaries”. So far, quantum cryptography has been mainly identified with the development of protocols for the distribution of a secret truly random key between two legitimate users, known as quantum key-distribution (QKD) protocols. Beyond QKD, quantum cryptography remains a largely unexplored area. One of the main ongoing projects at the and Technology group of IESL-FORTH [L1], is the design and development of cryptographic solutions, which rely on fundamental quantum-optical systems and processes, and offer security against quantum adversaries.

Electronic communications and transac - problem, which are considered to be both of the aforementioned mathemat - tions constitute one of the main pillars of hard to solve on classical computers, in ical problems efficiently. our society. The role of cryptography is the sense that there are no known effi - to ensure the stability of this pillar, by cient classical algorithms for solving Though a fully functional universal providing techniques for keeping infor - these problems for all integers in poly - quantum computer of the necessary size mation secure, for determining whether nomial time. Given, however, that the to break widely used cryptosystems is information has been maliciously non-existence of such algorithms has still far off in the future, there is a neces - altered, and for determining who never been rigorously proved, most of sity for ensuring the security, the authored pieces of information. To a the widely used public-key cryptosys - integrity and the authenticity of data large extent, modern public-key cryp - tems are susceptible to advances in that is encrypted or digitally signed tography relies on mathematical prob - algorithms or in hardware (computing today, and they have long lifetime. lems, such as the factorisation of large power). Indeed, Peter Shor has proved Unfortunately, the currently available integers and the discrete logarithm that a quantum computer could solve QKD systems become inefficient for

2,#+667",*-)*-."(0#"-*"#7&,)6 340AA"+/" '"()*+(" =.")0.0-1*+2*@2 NO2340AA"+/" =L! '"()*+(" 52 12 %0(". 53 13 M M *+ 0 &"0(?."5"+- ,-./ 1. 54 14 5. ;0-060("

%"#&'&()*&+,-)*--80,9."(0#"-*"#7&,)6 !"#$"# =,Q =.")0.0-1*+2*@2 340AA"+/" '"()*+(" =L! NO2340AA"+/" 52 12 %0(". 5 53 13 M . M *+,- /0 &"0(?."5"+- . 1 54 4 5. 6. Q* : P"( 6. 71. ! *+,-/0R2>?0+-?52(-0-"2*@2A1/4-2)0.05"-".1S":2 6#202("-2*@2?0+-?52<40AA"+/"(20."2+*+ D*.-4*/*+0A ;;"<1(1*+"<1(1*+ <"="(* ;(("1*

Figure1:Schematicrepresentationofaquantum-opticalEAP,whichreliesonachallenge-responsemechanism.Theenrolmentstageis performedoncebythemanufacturer,whiletheverificationstagetakesplaceeachtimetheholderofthePUKhastobeauthenticated.

ERCIM NEWS 112 January 2018 17 Special theme: Quantum Computing

large networks (many users), where the Currently, optical physical unclonable Links: establishment of many pairwise secret keys (PUKs) are considered to be the [L1] http://www.quantum- keys is needed, and the key manage - most promising candidates for the technology.gr/ ment remains a major problem. development of highly secure EAPs. [L2] http://gate.iesl.forth.gr/~nikolg Moreover, QKD alone does not address Such PUKs are materialised by an many other cryptographic tasks and optical multiple-scattering disordered References: functions, which are of vital importance medium, and they are considered to be [1] G. M. Nikolopoulos, E. Diamanti: in everyday life (e.g., authentication, unclonable, in the sense that their “Continuous-variable quantum non-repudiation, integrity, etc.). cloning requires the exact positioning authentication of physical (on a nanometre scale) of millions of unclonable keys”, Scientific Cryptographic research in our theoret - scatterers with the exact size and shape, Reports, 2017, 7, p. 46047. ical group started in 2007, and its which is considered to be a formidable [2] G. M. Nikolopoulos, T. Brougham: emphasis is on the design and the devel - challenge not only for current, but for “Decision and function problems opment of quantum cryptographic future technologies as well. Typically, a based on boson sampling”, primitives and protocols, which rely on PUK-based EAP relies on a challenge- Physical Review A, 2016, 94, p. fundamental quantum-optical systems response mechanism, in which the PUK 012315. and processes, and offer security against is interrogated by light pulses with ran - [3] G.M. Nikolopoulos: “Applications quantum adversaries [1-3]. Typical domly chosen parameters, and accept - of single-qubit rotations in quantum-optical systems are single ance or rejection of the PUK is decided quantum public-key cryptography”, photons, light in various quantum upon whether the recorded responses Physical Review A, 2008, 77, p. states, atoms, waveguides, and cavities. agree with the expected ones. Although, 032348. A description of our activities and in general, PUK-based EAPs are more related publications can be found at robust against cloning than conven - Please contact: [L1] and [L2]. tional EAPs, they are still vulnerable to Georgios M. Nikolopoulos emulation attacks when challenges per - IESL-FORTH, Greece Entity authentication (identification) is tain to classical light, and the verifica - [email protected] an important cryptographic task, in tion set-up is not tamper-resistant. To which one party (the verifier) obtains eliminate this vulnerability, in collabo - assurance that the identity of another ration with E. Diamanti (CNRS, party (the claimant) is as declared, Université Pierre et Marie Curie), we thereby preventing impersonation. have proposed a novel quantum-optical Most of the entity authentication proto - EAP in which a PUK is interrogated by cols (EAPs) used for everyday tasks randomly chosen non-orthogonal (e.g., transactions in automatic teller coherent quantum states of light (see machines), rely on dynamic challenge- Figure 1) [1]. The response of the PUK response mechanisms, which combine to a quantum state (challenge) is sensi - something that the claimant knows tive to the internal disorder of the PUK, (e.g., a PIN), with something that the which makes our protocol collision claimant possesses (e.g., a smart card). resistant, and robust against cloning. In such mechanisms, after the user types Moreover, on-going research shows that in the correct PIN, the smart card is the security of our protocol against an challenged with random numerical emulation attack relies on the laws of challenges, and the verifier checks if the quantum physics, which do not allow responses of the card are valid. unambiguous discrimination between Conventional EAPs are not totally non-orthogonal quantum states, while immune to card-cloning, while they are information gain cannot be obtained susceptible to emulation attacks, in without disturbing the quantum state which an adversary knows the chal - under interrogation. The proposed pro - lenge-response properties of the smart tocol can be implemented with current card (e.g., by hacking the database of technology, and its performance under challenge-response pairs), and his task realistic conditions is the subject of is to intercept each numerical challenge future theoretical and experimental during the verification stage, and send work. to the verifier the expected response.

18 ERCIM NEWS 112 January 2018 Quantum Lego: Graph States for Quantum Computing and Information Processing by Damian Markham (LIP6, CNRS - Sorbonne Université)

The massive global investment in quantum technologies promises unprecedented boosts for security, computation, communication and sensing. In this article we explore the use of so-called ‘graph states’ – a family of multipartite entangled states which act as ubiquitous resources for quantum information, are easily adapted for different tasks and applications, and can be combined in ways that fuses different utilities.

Quantum computing and quantum infor - together in the right way to achieve it. crucial – by combining protocols we get mation in general offer incredible bene - This capacity will be key in taking the functionality that is better than could be fits to our information society. Since the best advantage of quantum technologies achieved by any single protocol in iso - original discoveries of better than clas - in future quantum networks. Indeed, lation. But there are many exciting sical security in quantum key distribu - typically, more sophisticated applica - things still to do, and we’re still figuring tion and super-classical computational tions are built up by combining basic out how graph states can be used, in advantage, the field has exploded with protocols. several directions to push the limits of new possibilities for exploiting quantum quantum information. For example, encoding. The natural connection to graph theory graph states have proven a fertile test has proven an additional benefit of the space to understand the resources of In quantum cryptography we have seen graph state approach. It turns out that non-locality and contextuality and their new protocols for coin flipping, one can understand many properties of role in many quantum advantages. The quantum money and secure multiparty their use for quantum information – graphical notion of flow of information computation offering functionality and how computation flows, where infor - also lends itself to the study of exciting or security that is not possible classi - mation sits – entirely in terms of the new directions in quantum information cally. For certain communication tasks, underlying graph properties. This where the inherent ambiguity in causal quantum techniques provide an expo - allows many graph theory techniques to order has been shown to be yet another nential gap between what is possible be put into play to push more what source of quantum advantage. Recently, quantumly and classically. Even before quantum advantages can be had. we have also seen that graph states are universal quantum computers are born, Examples include graphical characteri - resources for the generation of quantum sub-universal devices will have a sation of where information sits in randomness, an almost generic resource plethora of applications from quantum secret sharing [1] and the application of in quantum information and key in our learning to simulation. In quantum random graph techniques for finding understanding in much of physics. metrology, quantum sensing provides optimal codes [2]. precision in measurements that would References: simply not be possible without uniquely Another consequence of their ubiquity [1] D. Markham, B. C. Sanders: quantum features. in quantum information is that there has “Graph states for quantum secret been a lot of effort to demonstrate them sharing”, Physical Review A 78.4 A remarkable family of multipartite experimentally. Indeed they represent (2008): 042309. entangled states acts as generic the cutting edge in what entangled [2] J. Javelle, M. Mehdi, S. Perdrix: resources for almost all these applica - states are prepared, with audacious “New Protocols and Lower Bounds for tions. Graph states are described in one experiments preparing graph states of Quantum Secret Sharing with Graph to one correspondence with a simple thousands of qubits. Experiments rou - States”, TQC 12 (2012): 1-12. graph, where vertices represent qubits (a tinely produce and control graph states [3] B. A. Bell, et al.: “Experimental two dimensional quantum system which of up to 10 qubits in different media in demonstration of a graph state is the basic quantum information unit) optics, atomics and ions. quantum error-correction code”, and edges represent a particular entan - Nature communications, 5, 2014 glement preparation. They are universal There are now several groups across resources for quantum computation, act Europe and the world working on Please contact: as codes for exploring graph state quantum informa - Damian Markham and are the entangled resource for many tion processing. In a series of works CNRS, LIP6, Sorbonne Université communication and cryptographic pro - with the groups of Mark Tame (Durban, [email protected] tocols. Perhaps their most compelling South Africa) and John Rarity (Bristol, strength is that they can be connected in UK) we have been pushing the quantum different ways to allow the utility of one Lego aspect, in particular. In one exper - function to be combined with another in iment we demonstrated a graph state a natural way. In this sense, these “graph protocol, which combined three dif - states” are like quantum Lego – decide ferent protocols to enable verified secret what you want to make and put them sharing [3]. This flexibility here proved

ERCIM NEWS 112 January 2018 19 Special theme: Quantum Computing

Graph Parameters and Physical Correlations: from Shannon to Connes, via Lovász and tsirelson

by Simone Severini (University College London)

Quantum information theory builds bridges between combinatorics, optimisation, and functional analysis.

Nowadays, it is theoretically predicted remarkable protocols for transmitting research direction has eventually rami - with paper and pencil, and experimen - information, including superdense fied into multiple routes. A brief tally demonstrated in the lab, that the coding and teleportation. account of the salient points follows. physical state of most multi-object quantum mechanical systems cannot be Two-player non-local games on graphs In 1956, Shannon defined the zero-error described by looking only at their indi - have been shown to be a particularly capacity of a communication channel as vidual components, and displays fruitful arena for rigorous research into the largest rate at which information can stronger-than-classical correlations. In the power of entanglement and, more be transmitted over the channel with virtue of this point, manipulating some generally, the correlations induced by error probability zero. The notion has of the components has a global effect on the axiomatic choices leading to their contributed to fuel a great amount of the whole system, even if the compo - different physical theories. In the pro - research in semidefinite programming nents are spatially separated – some - posed framework, two players, Alice and structural graph theory. Berge’s per - thing expressed by the currently over - and Bob, receive vertices of some fect graphs were remarkably motivated used and somehow unwelcoming graphs. Their task is to respond to ques - by the zero-error capacity and the Einstein’s “spooky action at distance”. tions without knowing each other’s ver - Lovász theta function was introduced as This phenomenon is known as “entan - tices; answers need to satisfy a certain an upper bound. When the parties can glement”. property as a function of the received share and locally manipulate entangled vertices. Depending on the type of cor - states, the analogue capacity has also When talking about entanglement, the relations displayed by the physical been proved to be bounded by Lovász average mathematical person tends to world of Alice and Bob, this situation theta, but to be exponentially larger in naturally refrain from alluding to the suggests a collection of new graph various single-shot and asymptotic halo of mysticism surrounding this term, parameters, which are captured by rich cases [1, 2]. Given the link between but to focus on the intricate structure of mathematical structures, and give an zero-error capacity and the problem of matrices and operators in composite occasion to generalise graph theory transmitting data over a broadcast Hilbert spaces. However, while entan - ideas to functional analysis. One of channel (e.g., coaxial cable or satellite), glement may appear as a flat conse - these quantities is now called quantum this result highlights a prospective quence of the tensor product used to chromatic number and it is known to be quantum advantage exploitable in real- combine systems as in wave-mechanics, loosely upper bounded by the more world communications. it also has the status of a legitimate and familiar chromatic number – when, in possibly fundamental physical quantity. the non-local game, answers need to be In 1976, Connes casually formulated a In truth, its freshly discovered applica - different for adjacent and equal for conjecture about a fundamental approx - tions range from information-theoreti - identical vertices. The quantum chro - imation property for finite von cally secure solutions for distributing matic number was the first quantum Neumann algebras – the Connes cryptographic keys to an assortment of graph parameter to be studied. Such embedding problem. Over time, many

Figure1:Twographson24verticesthatarequantumisomorphicbutnotisomorphic.TheconstructionisrelatedtotheFGLSSreductionfrom inapproximabilityliterature,aswellastheCFIconstruction.

20 ERCIM NEWS 112 January 2018 Somemembersofthe QuantumComputing, Information,and AlgebrasofOperators (QCIAO)collective (lecturersoftheLMS ResearchSchool Combinatoricsand OperatorsinQuantum InformationTheory, BelfastSeptember2016).

unexpected equivalent statements for familiar but powerful generalisation of the conjecture have emerged. A hier - graph colouring. This new technical References: archy beyond the quantum chromatic machinery became further consolidated [1] R. Duan, S. Severini, A. Winter: number has led to a novel reformulation via a quantum version of graph homo - “Zero-error communication via of the Connes conjecture, and generated morphism, a familiar but powerful gen - quantum channels, non- a fresh effort towards its solution [L1]. eralisation of graph colouring. The new commutative graphs and a quantum The new approach is centred on combi - type of homomorphism suggested Lovasz theta function”, IEEE natorial ideas lifted to the realm of oper - relaxations of graph isomorphism to Trans. Inf. Theory 59(2):1164- ator algebras. Moreover, extending this settings corresponding to various phys - 1174, 2013. mathematical landscape, hierarchies of ical theories. The findings of this line of [2] J. Briët, H. Buhrman, D. Gijswijt: quantum graph parameters associated to research are surprising. While fractional “Violating the Shannon capacity of correlations can be placed in the frame - isomorphism corresponds to sharing metric graphs with entanglement”, work of (tracial noncommutative) poly - some type of correlations stronger than PNAS 2013 110 (48) 19227-19232. nomial optimisation [L2] entanglement – hence, it gets an opera - [3] A. Atserias, et al.: “Quantum and tional interpretation, – quantum isomor - non-signalling graph In 1993, Tsirelson proposed some prob - phism is obtained by relaxing the isomorphisms”, ICALP 2017. lems concerned with deciding whether integer programming formulations for the axiomatic mathematical models of standard isomorphism to Hermitian Please contact: non-relativistic quantum mechanics, variables [3] (see Figure 1). Simone Severini where observers have operators acting University College London, UK on a finite dimensional tensor product The QCIAO Collective is an interna - +44 (0)20 3108 7093 space, and algebraic quantum field tional collaboration focused on the [email protected] theory, where observers have com - topics of this article [L5]. muting operators on a (possibly infinite dimensional) single space, produce the Links: same set of correlations. A 2016 break - [L1] https://arxiv.org/abs/1503.07207 through, based on geometric group [L2] https://arxiv.org/abs/1708.09696 theory, settled the “middle” Tsirelson [L3] https://arxiv.org/abs/1606.03140 problem, by observing that the set of [L4] https://arxiv.org/abs/1709.05032 (tensor-product) quantum correlations [L5] http://www.qciao.org/ is not closed [L3]. Interestingly, a suc - cessive alternative proof of this result makes use of quantum graph parameters [L4].

This new technical machinery became further consolidated via a quantum ver - sion of graph homomorphism, a

ERCIM NEWS 112 January 2018 21 Special theme: Quantum Computing

High-Speed Entanglement Sources for Photonic Quantum Computers

by Fabian Laudenbach, Sophie Zeiger, Bernhard Schrenk and Hannes Hübel (AIT)

Photonic quantum computers promise compact, user-friendly packaging. The building blocks of such an implementation comprise of sources for efficient production of photons with high purity. To increase the clock speed of the computation, such sources need to operate in the GHz range.

Although the current favourites for a erty for cryptographic application and general, multiplexing a large number of scalable quantum computer seem to be secure software distribution. such sources together will produce the super-conducting qubit implementations, necessary entangled state. One key a pure photonic-based solution should computing issue is that the photons produced must not be disregarded, since it offers room A scalable model for photonic quantum be indistinguishable, i.e. they must look temperature operation and small foot - computation is the so called one-way or all the same. Therefore, no correlations prints. As the standard approach to measurement based quantum computer in frequency, time or spatial modes are quantum computing is based on the [2]. In this approach, the actual compu - allowed between the photons. The quantum circuit model, a generalisation tation proceeds by single qubit meas - figure of merit for this property is the of the classical circuit, featuring logical urements only, on a large entangled purity, photons that are completely operations such as AND and XOR gates, resource state (cluster state). In a pho - uncorrelated have a high purity whereas single-qubit gates (e.g. rotations) and tonic implementation, the cluster state photons which correlate in a degree of two-qubit gates (e.g. controlled-NOT) is a set of large numbers of photons, freedom show a low purity. Figure 1 are required for universal quantum com - which are entangled to each nearest illustrates the absence of frequency cor - puting. This quantum circuit model is not neighbour. The advantage lies in the relation between two generated single very well suited to a photonic implemen - fact, that there is no difference between photons. tation, since the photon-photon interac - performing a single-qubit or two-qubit tion is negligible and a two-photon gate gate operation. In both cases, there will The GHz entanglement source is hence not feasible. be only measurements on single pho - The AIT-Austrian Institute of tons, a feat easy to achieve. Technology in Vienna successfully Computations over the quantum The main challenge lies of course in the developed a source for polarisation- internet creation of the entangled cluster state entangled photon pairs which can be With the near advent of quantum net - which sounds like a taunting task, espe - operated at a tunable repetition rate of works, consisting of distribution of cially for large states of several hun - up to 40 GHz to be used in cluster state single or entangled photons between dreds of qubits. Fortunately, it can be generation [3]. Our source is based on a users, a photonic quantum computer shown that heralded photon sources suf - Sagnac-interferometer, where a non - would have another advantage. Unlike fice to build a cluster state of any size. linear optical medium is pumped by two matter based quantum computers, which Of course efficiencies have to be pulsed laser beams from two diametri - require a photon/matter quantum inter - increased and losses minimised, but in cally opposed directions. By virtue of a face, a cluster-state quantum computer could be directly connected to a global “quantum internet”. Protocols which rely on a quantum link between user and quantum computer exist today. Imagine the scenario, in which quantum com - puters are still very large and expensive, and clients can only request computation time from a remotely located quantum computer centre which is not necessarily

trustworthy. Using the blind quantum 1 computing protocol [1], the client’s inputs, outputs and computation remain perfectly private, even to the operators of the quantum computer centre. In another algorithm, quantum enabled one-time programing, the program itself is encoded onto photons, sent to a clas - sical processing unit and executed. Since the qubits are destroyed during the 2 measurement process, the program will Figure1:Frequencycorrelationplot(jointspectralintensity)betweenthephotonsofthe only run once, a much sought-after prop - generatedpair.Thenearlycirculardistributionindicatesalowdegreeofcorrelations;in contrastamoreelongatedspectraldistributionwouldpointtoalowpurityofthephotons.

22 ERCIM NEWS 112 January 2018 higher detection efficiency but also operate at a higher time resolution. This will allow us to generate cluster states close to the maximal pulse rate of the laser at 40 GHz. While enlarging the number of photons in the cluster state, we will also strive to reduce the overall size of the experiment using photonic integration techniques.

Acknowledgements This work was funded by the Austrian Research Promotion Agency (Österre - ichische Forschungsförderungsgesell- schaft, FFG) through the project KVQ (No. 4642983).

Link: https://www.ait.ac.at/themen/optical- quantum-technologies

References: [1] S. Barz et al.: “Demonstration of Figure2:Left:Principleofentanglementgeneration.Dependingonthepathofthepumplaser Blind Quantum Computing”, (red),thevertical(green)andhorizontal(blue)photonsareexitingatdifferentports;Right: Science 335 (2012), 303 ExperimentalsetupofSagnacinterferometer,theredpumpbeamentersfromtherightside. [2] R. Raussendorf and H. J. Briegel: “A One-Way Quantum Compute”, PRL 86 (2001) 5188 nonlinear process called parametric Figure 2. The photons are generated [3] F. Laudenbach et al.: “Modelling downconversion (PDC), some of the with high purity and are highly entan - parametric down-conversion laser photons would decay within the gled (95%). yielding spectrally pure photon crystal and give birth to two photons pairs”, Opt. Exp. 24 (2016), 2712 with half the energy each. Entanglement Outlook emerges from the lost “which-way In the next step, we will simultaneously Please contact: information” after recombination of the pump three Sagnac-interferometers Hannes Hübel, Martin Stierle two paths: There is no possible way to (instead of one) in order to generate AIT Austrian Institute of Technology, tell which pump beam produced the larger cluster states. Moreover, we will Austria twin photons; therefore the two possible replace our avalanche photo diodes by [email protected], origins exist at the same time in a superconducting nanowire detectors [email protected] quantum superposition, as shown in which not only have significantly

A formal Analysis of Quantum Algorithms by Benoît Valiron (LRI – CentraleSupelec, Univ. Paris Saclay)

Moving from the textual description of an appealing algorithm to an actual implementation reveals hidden difficulties.

Between 2011 and 2013, I had the crete problems, with the task to imple - size of the input parameter grows, can opportunity to collaborate on the Pan- ment and estimate the resources neces - we asymptotically go faster with the use American project QCS (Quantum sary to effectively run them, in the of a than with purely Computer Science) devoted to the event of an actual quantum computer. classical means? It turns out that many implementation and analysis of quan - To this end we developed the quantum interesting problems have this property: tum algorithms. Researchers from dif - programming language Quipper [L1]. many fields ranging from big-data to ferent areas, ranging from physicists to chemistry and pharmaceutic could ben - theoretical computer scientists, were Quantum algorithms are not usually efit from the use of quantum algorithms. working on this project; I was at the lat - designed as practical handles but instead The natural question is how to con - ter end of this spectrum. We were given as tools to explore complexity bound - cretely implement these quantum algo - a set of seven, state-of-the-art quantum aries between classical and quantum rithms, estimate the required resources algorithms instantiated on some con - computation: for a given problem, as the and validate the implementation.

ERCIM NEWS 112 January 2018 23 Special theme: Quantum Computing

In the realm of classical computation, Quipper is an embedded language within the backbone of two projects aiming at implementing an algorithm implies the a host language. The chosen host lan - such goals. The European Itea3 project choice of a programming language to guage is Haskell. This modern, func - Quantex spanning across France, code it, a platform to run the resulting tional language features an expressive Netherlands and Germany gathers program, and a compiler that can turn type system making it easy to define and Atos/Bull, CEA/Leti, LORIA, LRI, the code into a program executable on enforce the specificities of circuit con - TUDelft, KPN, Siemens and Univ. this platform. The realm of quantum struction and manipulation. Tübingen. Quantex aims at developing computation is currently less developed: Parametricity is naturally obtained with the programming environment around several competing platform co-exists, the use of lists and list combinators. The Quipper and focuses on the compilation and in 2011, at the beginning of the QCS construction of a circuit is modelled as towards emulation of quantum compu - project, no scalable quantum program - printing on a particular kind of output tation. The recently started French ANR ming language even existed. channel: generating an elementary SoftQPro is centered on a formalisation instruction corresponds to writing it on of Quipper’s semantics toward certifica - From a programmer’s perspective, the channel. Within Haskell, this kind of tion, and on the development of a com - quantum computation is very close to side-effect can naturally be encapsulated pilation stack based on the graphical cal - classical computation. The main differ - within a type construct known as monad. culus ZX, envisioned as a more natural ence lies in the use of a special kind of This automatically outlaws various ill- intermediate representation than exist - memory with exotic properties: in par - defined programs, renders their coding ing proposals. ticular, quantum data is non-duplicable less error-prone while easing debugging. and reading quantum data is a proba - Link: [L1] https://kwz.me/hB6 bilistic operation. The interaction with Quipper has been used to code large the quantum memory is done by a algorithms and perform logical resource References: sequence of elementary operations that estimation: for a given set of parameters [1] A. S. Green, P. L. Lumsdaine, N. J. are summarised by what is known as a to the problem, what is the size of circuit Ross et al.: “Quipper: A Scalable quantum circuit. A quantum algorithm generated by the algorithm? The size can Quantum Programming Language”, mainly consists of the construction of be counted in the number of elementary, in Proc. of PLDI 2013, ACM such circuits, their execution on the logical operations and in the size of the SIGPLAN Notices, Vol. 48 Issue 6, quantum memory, and various classical required quantum memory footprint. pp. 333-342, 2013. pre- and post-processing. The analysis shows that for naive imple - [2] B. Valiron, N. J. Ross, P. Selinger et mentations, these numbers can be quite al.: “Programming the Quantum This hints at the main required design large, yielding unrealistic circuits. This Future”, Communications of the choice: a quantum programming lan - analysis tends to show that producing ACM, Vol. 58 No. 8, pp. 52-61, guage is primarily a circuit-description usable algorithms requires more than 2015. language. However, quantum algorithms concentrating on complexity analysis. [3] A. Scherer, et al.: “Concrete are not simply fixed, static quantum cir - Nowadays, it is possible to do so: with resource analysis of the quantum cuits. Unlike classical circuitry such as the help of programming languages such linear-system algorithm used to FGPAs, quantum algorithms describe as Quipper, a quantum algorithm compute the electromagnetic families of quantum circuits: the circuit designer can effectively analyse and tune scattering cross section of a 2D depends on the size and shape of the its algorithm for concrete use-cases. target”, Quantum Information input data. A quantum programming lan - Processing 16:60, 2017. guage therefore has to account for this The advent of modern quantum pro - parametricity. Quipper has been gramming languages clears a path Please contact: designed from the ground up with these towards the design of quantum compila - Benoît Valiron, LRI – CentraleSupelec, aspects in mind. Following a successful tion stacks and tools for certification of Univ. Paris Saclay, France trend in domain specific languages, quantum programs. Quipper is currently [email protected]

diagram transformations Give a New Handle on Quantum Circuits and foundations

by Aleks Kissinger (Radboud University)

String diagrams provide a powerful tool for uncovering hidden algebraic structure in quantum processes. This structure can be exploited to optimise quantum circuits, derive fault-tolerant computations, and even probe the foundations of physics.

Quantum computers have the ability to physics, chemistry, and computer sci - memory, computational power, and tol - reshape the modern world, offering huge ence. However, the existing and erance to noise, making it crucial to computational speed-ups for a wide emerging quantum computational develop sophisticated techniques for variety of problems in mathematics, devices face stringent limitations in optimising the software which drives

24 ERCIM NEWS 112 January 2018 families of circuits are capable of pro - ducing any quantum computation imag - inable (or, in the case of Clifford+T, approximating it to arbitrarily high pre - cision). Thus, a complete algebraic characterisation of equivalence for these circuits is a major breakthrough.

The ZX-calculus—including the exten - sions proposed by the Oxford and LORIA groups—is based on string dia - grams, which can be seen as a sort of Figure1:Decomposingquantumgatesintomoreprimitivepiecescalled“spiders”. generalisation of quantum circuits. They first appeared in the work of Penrose in the 1970s, and since then these systems. Quantum circuits have practical language for reasoning about have been applied to a broad range of become the de facto “assembly lan - quantum circuits, as well as other qubit- applications in physics and computer guage” for quantum software. They based models of computation, such as science, including tensor networks in describe computations in terms of a measurement-based quantum computa - high-energy physics, signal-flow series of primitive operations, called tion. The ZX-calculus works by decom - graphs, electrical and electronic cir - quantum gates, performed on a register posing quantum gates into even more cuits, computational linguistics, and of quantum bits (qubits), which is then primitive components, called “spiders” concurrent computation. Earlier this measured to give a result. While (Figure 1). These basic pieces satisfy a year, Coecke and Kissinger published a quantum gates are useful as building small number of algebraic laws, which textbook which gives a comprehensive blocks for computations, they lack a in turn yield a great deal of power. For introduction to quantum theory, well-understood algebraic structure, example, the four laws shown in Figure quantum computation, and quantum making it difficult to understand when 2 suffice to transform any two equiva - foundations purely in the language of two circuits are equivalent (i.e., describe lent Clifford quantum circuits into the string diagrams [3]. the same computation) or to transform same circuit. Clifford circuits are a one circuit into another for the sake of well-studied class of circuits which can String diagrams not only provide a optimisation or fault-tolerance. be simulated efficiently on a classical unique and intuitive way to introduce computer, so it may not be too sur - the core concepts of quantum theory, In 2008, researchers at Oxford prising that the ZX-calculus gives an but also a dramatically different way of University produced a unique solution easy, algebraic handle on circuit equiva - working with quantum mechanical to this problem: the ZX-calculus. lence. However, what is surprising is processes. Within the diagrammatic Originally developed as an abstract that groups in Oxford and LORIA (a approach to quantum theory (a.k.a. method for studying the behaviour of research unit affiliated with ERCIM “Categorical Quantum Mechanics”, see complementary observables in quantum member Inria) have shown in recent link [L2]), concepts such as connec - mechanics (indeed the ‘Z’ and ‘X’ refer months that an extension to these rules tivity, composition, and interaction take to the complementary Pauli observables can decide equality for Clifford+T cir - centre stage, whereas concrete Hilbert- of the same name), the ZX-calculus cuits [1] and even a fully universal space calculations are secondary. For quickly showed itself to be a useful family of quantum circuits [2]. These example, foundational questions around

Figure2:TherulesoftheZX-calculus,whichsufficetoderiveallequationsthatholdbetweenCliffordcircuits.

ERCIM NEWS 112 January 2018 25 Special theme: Quantum Computing

Figure3:Quantomatic,aproofassistantfordiagrammaticreasoning.Here,itiscomputingatransversalimplementationofaCCZgatewithina quantumerrorcorrectingcode(namely,the[[8,3,2]]colourcode;seelinks[L1]and[L3]formoredetails).

quantum non-locality and quantum The teams in Nijmegen, Oxford, Links: causal structure can be posed in terms of LORIA, and Strathclyde, with the help [L1] http://quantomatic.github.io whether a diagram decomposes in a cer - of collaborators at LRI, Durham, and [L2] http://cqm.wikidot.com tain way across time and space, and the UK’s NQIT Quantum Hub, are now [L3] https://kwz.me/hB7 what consequences that decomposition aiming to produce fully automated tech - has on our observations. niques for circuit optimisation, scale up References : to large computations on hundreds of [1] E. Jeandel, S. Perdrix, R. Vilmart: More pragmatically, quantum algo - logical qubits (with error correction), “A Complete Axiomatisation of the rithms and communication protocols and develop new kinds of transforma - ZX-Calculus for Clifford+T can be proven correct using diagram tion procedures to work within the con - Quantum Mechanics”, Preprint, transformation rules, even in cases straints of first-generation quantum arXiv:1705.11151. 2017. involving far too many qubits for con - hardware, such as limited topologies for [2] K. Ng, Q. Wang: “A universal crete calculation. To assist with pro - qubit interactions and distinguished completion of the ZX-calculus”, ducing, checking, and sharing these gate- vs. memory-optimised physical Preprint, arXiv:1706.09877. 2017. proofs, researchers at Radboud qubits. [3] B. Coecke, A. Kissinger: “Picturing University, University of Strathclyde, Quantum Processes: A First Course and Oxford have developed a tool called in Quantum Theory and Quantomatic (Figure 3). Quantomatic is Diagrammatic Reasoning”, a “proof assistant for diagrammatic rea - Cambridge University Press. 2017. soning”. Using a combination of human- guided diagram transformations and Please contact: automated rewrite strategies (program - Aleks Kissinger mable in a Python-based strategy lan - iCIS Radboud Universiteit, The guage), it is possible to perform a variety Netherlands of tasks in Quantomatic, such as opti - [email protected] mising small to medium-sized quantum circuits, verifying multi-party communi - cation protocols, and computing encoded logical operations within cer - tain families of quantum error correcting codes.

26 ERCIM NEWS 112 January 2018 Quantum Computers and their Software Interfaces by Peter Mueller, Andreas Fuhrer and Stefan Filipp (IBM Research – Zurich)

Scientific groups in industry and academia have made enormous progress in the implementation of first quantum computer prototypes. IBM’s quantum experience with five and 16 qubits are already publicly accessible in the cloud. Three “standard” software interfaces are available. Client systems with 20 qubits ready for use and the next-generation IBM Q system is in development with the first working 50 qubit processor.

Quantum computers are expected to Figure1:Opencryogenic solve problems which are intractable to systemfora50qubit classical information processing. Built quantumcomputingdevice. on the principles of quantum mechanics, Thedeviceishiddeninside they exploit the complex and fascinating theshieldingtubeattachedat laws of nature at the molecular and bottomcenter.Thecopper atomic level, which usually remain structurestotheleftand hidden from view. By harnessing this rightoftheshieldingtube quantum behaviour, quantum computing containparametricquantum can run new types of algorithms to amplifierstoreadout process information in extremely large informationfromthequbits. state spaces. These algorithms may one day lead to revolutionary breakthroughs in materials simulation, the optimisation of complex manmade systems, and in machine learning.

Quantum computers are based on qubits, which operate according to two key principles of quantum physics: superpo - sition and entanglement. Superposition means that each qubit can represent both a “1” and a “0” at the same time [1]. It leads to what is called “quantum paral - lelism”, an effect that allows an n-qubit register to store all 2n possible states at the same time. Entanglement means that the state of multiple qubits can be corre - lated with each other; that is, the state of mation processing and to couple qubits in Figure 2 (left and middle). Currently, one qubit (whether it is a “1” or a “0”) to I/O ports (c) to manipulate, write and 5-qubit and 16-qubit devices are pub - depends on the state of one or more read the qubits. All the qubit manipula - licly available to run user programs other qubits and the qubit can no longer tions and readout processes are con - from the cloud. Alternatively, a cloud- be described separately. Using the prin - trolled by microwave pulses in the based simulator software which handles ciple of entanglement creates additional range of 4 to 8 GHz. up to 20 qubits can be chosen. processing capabilities which are not Examples of the three different pro - available on classical processors. The “IBM Q experience” provides a gramming interfaces are depicted in platform to explore basic quantum cir - Figure 3. The graphical user interface is IBM’s quantum devices are built on cuits on real quantum devices as shown the simplest way to access the real superconducting Josephson junction technology which requires cryogenic temperatures as depicted in Figure 1. The quantum bits are anharmonic LC resonators, known as qubits [2], where the non-linear inductor (L) is implemented by a Josephson junction. The left outer device in Figure 2 shows five square shaped qubits (a). The mean - dering connections (b) are coplanar Figure2:Anexperimental5-qubit(left)and16-qubit(mid)deviceasusedinthepublicly waveguides (CPW). They are used to available“IBMQexperience”andapackageasusedfortheupcoming20/50qubitdevices couple qubits with each other for infor - (right).Forfurtherdetailsseetext.

ERCIM NEWS 112 January 2018 27 Special theme: Quantum Computing

form to explore basic quantum circuits on real quantum processors based on superconducting qubits. It has been used by more than 60,000 users and has already generated more than 35 third- party research publications on applica - tions of quantum computing. On the software side, a growing number of developers are using . To sum - marise, the time is now to join forces, execute and continue to make key breakthroughs on this revolutionary journey of information technology.

Links: http://ibm.com/ibmq/ http://www.qiskit.org/ http://math.nist.gov/quantum/zoo/

References: Figure3:Thethree“IBMQexperience”webinterfaces–graphicaluserinterface(top), [1] M. A. Nielsen and I. L. Chuang, OpenQASMassembler(bottomleft)andtheQuantumInformationSoftwareKit(bottomright)– “Quantum Computation and giveaccesstotheexperimentalfiveand16qubitcircuits. Quantum Information”, 10th edt., Cambridge University Press, 2010. [2] J. Koch et al., “Charge-insensitive quantum hardware using a web open-access interface allows a user- qubit design derived from the browser. OpenQASM [3] is a low level application to integrate the quantum Cooper pair box”, Phys. Rev. A 76, hardware interface which enables com - processor as an accelerator, through the 042319, 2007. patible software stacks to be built. For cloud. [3] A. W. Cross et al., “Open Quantum more advanced developments, the Assembly Language”, Quantum Information Software Kit On the application side, calculations of [arXiv:1707.03429], 2017. (QISKit) is an open-access programing e.g., the ground state energies of small interface which provides highest func - molecules such as H2, LiH and BeH2 Please contact: tionality for working with both, the real have been reported. Further examples Peter Mueller quantum processors and various soft - are shown in the links given below. The IBM Research – Zurich, Switzerland ware based quantum simulators. The “IBM Q experience” provides a plat - pmu@zurich..com

Chevalley-Warning theorem in Quantum Computing

by Gábor Ivanyos and Lajos Rónyai (MTA SZTAKI, Budapest)

Effective versions of some relaxed instances of the Chevalley-Warning Theorem may lead to efficient quantum algorithms for problems of key practical importance such as discrete logarithm or graph isomorphism.

The Theory of Computing Research nent special cases as the task of com - and best known example of this is Group of the Informatics Laboratory at puting discrete logarithms and the ques - finding periods for functions defined on MTA SZTAKI has expertise in algebraic tion of finding isomorphisms of graphs. the integers. One of the greatest suc - aspects of quantum computing, The object we are given is a function f cesses of quantum algorithms, Shor’s including quantum algorithms for alge - defined on a large finite group G and we method for factoring integers, is based braic and arithmetical problems, as well are looking for the subgroup H con - on finding such a period. Computing dis - as application of algebraic methods as sisting of all elements h for which f(xh) = crete logarithms in various settings are is ingredients of quantum algorithms. f(x) for every x from G. In other words, H also an instance of the hidden subgroup Some of our projects aim at discovering is the group of elements whose action problem over abelian groups. hidden algebraic structures, e.g., symme - leaves f invariant. (We remark note that tries of certain objects. A main example in most cases, we further require that f is The graph isomorphism problem can be is the so-called “hidden subgroup such that f(x) = f(y) if and only if y = xh cast as an instance of the hidden sub - problem”, which includes such promi - for some h from H.) Perhaps the simplest group problem over a noncommutative

28 ERCIM NEWS 112 January 2018 group G. In contrast with the commuta - degree and toward hidden subgroup tive case, for which efficient quantum algorithms in some more complex algorithms are known, the complexity of groups. the noncommutative hidden subgroup problem has remained open even for References: certain groups that are very close to The condition that allowed us a method commutative ones. Among the few posi - running in polynomial time is that the [1] K. Friedl, et al: “Hidden translation tive results in this direction, we mention number of variables, compared to the and translating coset in quantum our polynomial time algorithm, devel - number of equations and the degree d, is computing” SIAM Journal on oped in a joint work [1], which finds sufficiently large. (Here by polynomial Computing 43 (2014), pp. 1-24. hidden subgroups in a fairly wide class time we mean time bounded by a func - [2] T. Decker, et al.: “Polynomial time of groups in which the order of the ele - tion polynomial in the bit size of the quantum algorithms for certain ments is bounded by a constant. The array of the coefficients, which is the bivariate hidden polynomial overall progress is much more modest number of equations, m, times the problems”, Quantum Information even in “so-called two-step solvable number of variables, n, times the loga - and Computation 14 (2014), pp. groups” (these are in a certain sense rithm of the size of the base field). In the 790-806. composed of two commutative groups) quantum setting in which our algorithm [3] G. Ivanyos, M. Santha: “Solving in which elements of larger order are is applied, the degree is essentially the systems of diagonal polynomial present. degree of the hidden polynomial map equations over finite fields,” and the number of equations is related to Theoretical Computer Science 657 With our collaborators we found [2] that the dimension of the underlying spaces, (2017), pp. 73-85. the hidden subgroup problem for a sub - while we are allowed to choose the class of such groups can be further gen - number of variables. (Note however, Please contact: eralised to another class of problems that the system is not required to be Gábor Ivanyos and Lajos Rónyai regarding hidden algebraic structure. In sparse, the n times m array of the coeffi - MTA-SZTAKI, Hungary this class, the hidden object is a polyno - cients can be arbitrary.) [email protected], mial map between vector spaces over a [email protected] finite field. Certain hidden subgroup Observe that without any assumption on problems can be formulated as hidden the number n of variables, already over polynomial map instances (there is a the field consisting of three elements the reduction in the other direction as well, quadratic case of the problem becomes but this results in a bigger hidden sub - NP-hard. This can be shown by a modi - group problem). A simple illustrative fication of the standard reduction of SAT example of a hidden polynomial map is to Subset sum . (In fact, that case of the as follows: let f(X) be an unknown uni - problem is just finding a zero-one solu - variate polynomial of constant degree. tion of the corresponding linear system.) We have access to a quantum oracle On the other hand, from the Chevalley- which returns E(Y 2-f(X) ) for given pairs Warning Theorem it follows that if n > (X,Y) . Here E is an unknown injective md , then our system always has a non - encoding of the field. The task is to trivial solution. Then an interesting determine f (up to constant term). We question arises: how hard is it to find a developed [2,3] a polynomial time nontrivial solution? There is some evi - quantum algorithm for finding such dence (such as the above mentioned hidden polynomial maps under the hardness result) that this question is too assumption that they have constant difficult when the number of variables is degree. close to the Chevalley-Warning bound md . For this reason, we look for an effi - One of the critical ingredients of our cient solution for relaxations in which n quantum algorithm is a classical algo - is substantially larger than this bound. rithm that under certain conditions finds First, it is worth noting that by a simple a nontrivial solution of a system of poly - and natural recursive algorithm, it is nomial equations of a very special kind, easy to find a solution in polynomial for which the basic and famous time, when the number n of variables is Chevalley-Warning theorem of number greater than a function like d raised to theory ensures the existence of a non - the m-th power. This method is useful trivial solution. Our system is obtained when m is constant. What can we say from a system of homogeneous linear when d is kept constant? For this variant equations by replacing each variable by of the problem, we have developed a its d-th power where d is a fixed positive much more sophisticated algorithm [3]. integer: This result is probably not optimal and an improvement could be a first step toward quantum algorithms for finding hidden polynomial maps of higher

ERCIM NEWS 112 January 2018 29 Special theme: Quantum Computing

Hardware Shortcuts for Robust Quantum Computing

by Alain Sarlette (Inria) and Pierre Rouchon (MINES ParisTech)

Despite an improved understanding of the potential benefits of quantum information processing and the tremendous progress in witnessing detailed quantum effects in recent decades, no experimental team to date has been able to achieve even a few logical qubits with logical single and two qubit gates of tunably high fidelity. This fundamental task at the interface between software and hardware solutions is now addressed with a novel approach in an integrated interdisciplinary effort by physicists and control theorists. The challenge is to protect the fragile and never fully accessible quantum information against various decoherence channels. Furthermore, the gates enacting computational operations on a qubit do not reduce to binary swaps, requiring precise control of operations over a range of values.

The major issue for building a quantum the fact that new degrees of freedom blocks towards more complex error cor - computer is thus to protect quantum and additional operations also carry rection strategies. Such hardware effi - information from errors. The quantum new possibilities for inducing errors or ciency is typically achieved when many counterpart of error correcting codes cross-correlations, can degrade the energy levels are associated to one and provides an algorithmic solution, based overall performance. the same physical degree of freedom. A on redundant encoding, towards scaling prototypical example is the “cat qubit”, up the precision in this context. In the QUANTIC lab at Inria Paris [L1], where information about a single qubit is However, it first requires technology to we are pursuing a systems engineering redundantly encoded in the infinite- reach a threshold at which the hardware approach to tackle this issue by drawing dimensional state space of a harmonic achieves the following on its own: inspiration from both mathematical oscillator, in such a way that the domi - • all operations, in any big system, must control theory and the algorithmic error nant errors reduce to the single photon already be accurate to a very high pre - correction approach (Figure 1). loss channel; and this error is both infor - cision; mation-preserving and non-destructively • the remaining errors follow a particu - A first focus of the group is to design a detectable. This overcomes the need for lar scaling model, e.g., they are all hardware basis where a large Hilbert tracking multiple local error syndromes independent in a network of redun - space for redundant encoding of infor - and applying according corrections in a dant qubits. mation comes with just a few specific coupled way, as would be the case for a decoherence channels. Concentrating logical qubit encoded on a network of Under these conditions, adding more efforts on rejecting these dominant noise -1/2 subsystems. An implementation physical degrees of freedom to encode sources allows more tailored and simple of this principle in superconducting cir - the same quantum information keeps stabilising control schemes to be applied cuits has recently achieved the first leading to better protection. Otherwise, to improve the quality of these building quantum memory improvement thanks to active error correction.

A second point of attention is the speed of operations. Indeed, quantum opera - tions take place at extremely fast timescales — e.g., tens of nanoseconds in the superconducting circuits favoured by several leading research groups. This leaves little computation time for a dig - ital controller acting on the system. Conversely, it hints at the fact that a smartly designed quantum system could stabilise fast on its own onto a protected subspace, possibly in combination with a few basic control primitives. Such autonomous stabilisation can be approached by a “reservoir engineering” Figure1:Generalschemeofourquantumerrorcorrectionapproachvia“hardwareshortcuts”. strategy, where conditions are set up for Thetarget“quantummachine”isfirstembeddedintoahigh-dimensionalsystemthatmayhave dissipation channels to systematically strongdecoherence,butonlyalongafewdominatingchannels(red).Then,apartofthe push the system into a desired direction. environmentisspecificallyengineeredandcoupledtoitinordertoinduceastronginformation- This can enact e.g., fast reset of qubits, stabilisingdissipation(blue).Thisentailsstabilisingthetargetquantummachineintoatarget or error decoding and correction, all in subspace,e.g.,“four-leggedcats”ofaharmonicoscillatormode;andpossibly,replacingthe one fast and continuously operating measurementoferrorsyndromesandconditionalaction,byacontinuousstabilisationofthe (“pumping”) hardware without any error-lesscodewords.Developmentofhigh-ordermodelreductionformulasisneededtogo algorithmic gates. Such “reservoir engi - beyondthisfirst-orderideaandreachtheaccuraciesenablingscalablequantuminformation neering” also plays an important role in processing. effectively isolating quantum subsys -

30 ERCIM NEWS 112 January 2018 tems with particularly concentrated improved and scaled up to reach the References: error channels, thus in creating the con - requirements 1. and 2., since every con - [1] M. Mirrahimi, Z.Leghtas et al: ditions of the previous paragraph. trol scheme can only be as precise as its “Dynamically protected cat-qubits: underlying model. We are therefore a new paradigm for universal quan - This brings us to the more mathematical launching a concrete effort towards tum computation”, New Journal of challenges. The design of engineered high-precision model reduction for - Physics, 2014. reservoirs, such as the singling out of a mulas, together with the supercon - [2] N. Ofek et al.: “Extending the life - “quantum machine” from its environ - ducting circuit experimentalists to iden - time of a quantum bit with error ment, is based on approximations where tify typical needs, but with a general correction in superconducting cir - weak couplings are neglected with scope in mind. Preliminary work cuits”, Nature, 2016. respect to dominant effects. These already shows the power of higher- [3] J. Cohen: “Autonomous quantum approximations enable tractable sys - order formulas in identifying design error correction with superconducting tems engineering guidelines to be opportunities (Figure 1), paving the qubits”, PhD thesis, ENS Paris, 2017. derived — rather than having to treat a way to improved “engineered” hard - single big quantum bulk, whose proper - ware for robust quantum operation. Please contact: ties are one target of the quantum com - Alain Sarlette, Pierre Rouchon, Inria puters themselves. Current designs are Link: [email protected], based on setting up systems with first [L1] [email protected] dominant local effects. This has to be https://www.inria.fr/en/teams/quantic + 33 1 80 49 43 59

Quantum Gates and Simulations with Strongly Interacting Rydberg Atoms by David Petrosyan (IESL-FORTH)

In order to develop functional devices for quantum computing and analogue and digital quantum simulations, we need controlable interactions between the physical systems representing quantum bits – qubits. We explore strong, long-range interactions between atoms excited to high- lying Rydberg states to implement quantum logic gates and algorithms and to realise quantum simulators of various spin-lattice models to study few- and many-body quantum dynamics.

Quantum systems with many degrees of rithms for certain mathematical tasks, quantum information processing tasks freedom, such as those composed of such as search of unstructured database [2]. many interacting subsystems or parti - (Grover) or integer factorisation (Shor), cles, are notoriously hard to simulate on are polynomially or exponentially more The interatomic interactions are con - classical computers. This is because the efficient than the best known classical trollable by lasers that can excite and Hilbert space for the quantum states of algorithms for the same tasks. de-excite the atoms from the non-inter - the composite system is exponentially acting ground state to the strongly inter - large in the system size, while quantum Cold atoms trapped in optical lattices or acting Rydberg state on demand. This correlations, or entanglement, between arrays of microtraps represent a scalable has led to proposals to implement the constituent subsystems often pre - architecture to realise quantum com - quantum logic gates using the switch - clude factorisation of the problem into puters as well as analogue and digital able interactions, or interaction-induced smaller parts. This has led Richard quantum simulations of many-body excitation blockade, between the atoms Feynman to suggest, back in 1981, to dynamics of various spin lattice models. representing qubits. There have been simulate quantum physics with quantum Neutral atoms in the lower electronic several experimental demonstrations of computers, or universal quantum simu - states interact very weakly with each the Rydberg quantum gates, but the lators [1] composed of many quantum other, but when excited to the Rydberg fidelity of operations has so far been two-level systems, like spin-1/2 parti - states, their interaction can be very below the threshold value ~0.9999 for a cles arranged in a lattice, with appropri - strong over large distances. Rydberg scalable fault-tolerant quantum compu - ately controlled couplings between states are highly excited states with the tation. We study the performance of them. In principle, any many-body atomic placed on an orbit that quantum algorithms under realistic Hamiltonian dynamics can be decom - is far from the ionic core. Due to weak experimental conditions involving posed into small time-steps and finite- binding of the electron to the ion, the noise and imperfections (see Figure 1). range interactions between the qubits. atoms in the Rydberg states are easily We devise novel schemes for high- This idea has then developed into a gen - polarisable. The resulting long-range fidelity quantum logic gates using eral purpose quantum computer which is interactions between the Rydberg atoms blockade and resonant exchange inter - suitable not only for digital simulations makes them uniquely suited for real - actions between the Rydberg state of physical systems, but also for ising strongly-interacting many-body atoms. This work is being done in col - quantum computations. Quantum algo - systems and for implementing various laboration with the theory group of

ERCIM NEWS 112 January 2018 31 Special theme: Quantum Computing

Klaus Mølmer at the Aarhus University Arrays of trapped atoms excited on logue quantum simulations. In addition and the experimental group of Mark demand to the Rydberg states by lasers to switchable interactions, relaxations Saffman at the University of can realise various spin-lattice models and energy dissipation can be intro - Wisconsin—Madison, USA. that can serve for both digital and ana - duced in this system in a controlled way. We study the dynamics of few- and many-body quantum systems using such Rydberg quantum simulators. As an example, in Figure 2 we show simu - lations of quasi-crystals of Rydberg excitation as observed in the experi - ments with laser driven atoms in optical lattices [3]. We collaborate with the theory group of Michael Fleischhauer at the University of Kaiserslautern, Germany, and part of this work is being done within the EU H2020 FET Proactive project RySQ (Rydberg Quantum Simulations) which involves many leading theory and experimental groups in Europe.

Finally, strong transitions between the Rydberg states of atoms in the microwave frequency range enable Figure1:GroversearchalgorithmwithRydbergblockadedatoms.Left:Levelschemeofthe their efficient coupling to electrical cir - registeratomsinteractingwithamicrowavefieldonthequbittransitionandwitharesonant cuits involving superconducting qubits laserfieldonthetransitiontotheRydbergstate.AtomsinRydbergstatesinteractwitheach and resonators. Superconducting qubits

otherviaastrong,long-rangepotentialV aa whichsuppressesRydbergexcitationofallbutone can realise fast quantum gates but are atomatatime.Right:ProbabilitiesofmeasuringcorrectoutcomesoftheGroversearchversus less suitable for storage of quantum numberofiterations,forN=2,3,4digitquantumregister,without(top)andwith(bottom)decay information. Coupling atoms to solid- oftheRydbergstate. state systems permits the realisation of hybrid quantum systems composed of different components with complemen - tary functionalities including quantum information processing, storage and conversion to optical photons for long- distance quantum communication. This is another direction of our research on quantum computation and simulations with Rydberg atoms.

Links: http://www.quantum-technology.gr/ http://qurope.eu/projects/rysq/

References: [1] S. Lloyd: “Universal Quantum Simulators”, Science, Vol. 273 (5278), pp. 1073-1078, 1996. [2] M. Saffman, T. G. Walker, and K. Mølmer, “Quantum information with Rydberg atoms”, Rev. Mod. Phys. Vol. 82, pp. 2313-2363, 2010. [3] P. Schauß et al., “Crystallization in Ising quantum magnets”, Science Vol. 347 (6229), pp. 1455-1458, 2015.

Figure2:Realisinginteractingspinlatticemodelswithlaserdrivenatoms.Upperpanel Please contact: illustratesanopticallatticepotentialforgroundstateatomsandlasercouplingtothestrongly David Petrosyan interactingRydbergstate.LowerpanelshowsspatialconfigurationofsixRydbergexcitations, IESL-FORTH, Greece withtheaxialP( ρ)andangularP( φ)probabilitydistributions,ina2Ddiskshapedlatticeof400 +30-2810 39 1131 atomsdrivenbyaresonantfield. [email protected]

32 ERCIM NEWS 112 January 2018 Control of Quantum Systems by broken Adiabatic Paths by Nicolas Augier (École polytechnique), Ugo Boscain (CNRS) and Mario Sigalotti (Inria)

The dynamics of a quantum mechanical system is described by a mathematical object called Hamiltonian. The possible results of an energy measure are known as the “eigenvalues” (or energy levels) of the Hamiltonian. After an energy measure, the system collapses into a particular state called the eigenstate corresponding to the measured energy. Adding corners to adiabatic paths can be used to generate superpositions of eigenstates with simple and regular control laws.

The goal of quantum control is to design one from another and in particular they the controls have more than one degree external fields (the controls) whose do not intersect. The theory can be of freedom. action induces a prescribed transition extended to systems presenting eigen - between two states of a quantum system. value intersections. When the intersec - In the general case one looks at the sin - tion of two eigenvalues is transversal gular set, i.e., the set of all control values A widely used technique is based on the and the initial condition corresponds to for which the Hamiltonian has degen - principle of adiabatic evolution: If the the lower eigenvalue then, after the erate eigenvalues. The applicability of initial condition is an eigenstate of the intersection, the trajectory adiabatically the adiabatic strategy relies on the fact controlled Hamiltonian (i.e., the follows the eigenstate corresponding to that the singular set does not disconnect Hamiltonian including the external field) the upper eigenvalue. the set of available control parameters. and the variation of the external fields is Such a disconnection is very rare, how - slow, then the state of the system stays One is typically interested in adiabatic ever, in the sense that for a generic close to the instantaneous eigenstate paths for which the controls start and system the singular set has codimension (namely, the eigenstate of the controlled end at zero. Such controls may be used 3 for complex Hamiltonians and codi - Hamiltonian with frozen time). to induce nontrivial transitions only if mension 2 for real ones. there exist paths in the space of controls Adiabatic evolution works as described that are not passing back and forth Figure 1 shows the eigenvalues of a 3- when the energy levels of the controlled through the same intersections. This is level quantum system driven by two Hamiltonian satisfy a gap condition, why such an adiabatic setup makes controls in the well-known STIRAP which means that they stay away from sense for control purposes only when configuration. In this case there are two eigenvalue intersections: one between the first and the second levels (on the axis where the first control is zero) and one between the second and the third levels (where the second control van - ishes). Notice that the resulting adia - batic strategy follows the famous counter-intuitive pattern, meaning that one first activates the control associated with the transition between the second and the third energy levels and then the one associated with the transition between first and the second ones, with an overlap between the two pulses [1].

The technique explained above pro - vides a simple strategy to induce a tran - sition between any two eigenstates of the free Hamiltonian, under the assump - tion that all energy levels intersect.

In collaboration with Francesca Chittaro (Toulon University) and Paolo Mason (CNRS, CentraleSupelec) we have developed a technique to create superpositions between eigenstates using broken adiabatic paths [2].

Figure1:Theeigenvaluesofa3-levelquantumsystemdrivenby2controlsintheSTIRAP The principle of our approach is that if configuration. an adiabatic path reaches a point in the

ERCIM NEWS 112 January 2018 33 Special theme: Quantum Computing

The project is supported by the ERC starting grant GECOMETHODS (2010- 2016) and by the French ANR projects GCM (2009-2013). Extensions and algo - rithmic refinements are under study in the framework of the Inria team CAGE thanks to the support of the French ANR project QUACO (2017-2021).

References: [1] C.E. Carroll and F.T. Hioe: “Analytic solutions for three-state systems with overlapping pulses”, Phys. Rev. A, vol. 42, pp. 1522- 1531, 1990. [2] U. Boscain et al.: “Adiabatic control of the Schrödinger equation via conical intersections of the eigenvalues”, IEEE Trans. Autom. Control, vol. 57, pp. 1970-1983, Aug. 2012. [3] U. Boscain et al.: “Approximate controllability, exact controllability, Figure2:Abrokenadiabaticpathinducingatransitionfromthefirstenergyleveltoa and conical eigenvalue intersections superpositionofthefirstandthethirdenergylevels. for quantum mechanical systems”, Commun. Math. Phys, vol. 333, pp. 1225-1239, Feb. 2015. singular set and makes there a corner, tion from the first energy level to a super - then the state splits partially on the position of the first and the third energy Please contact: lower energy level and partially on the levels. Given the desired population Nicolas Augier, CMAP, École upper one. Modulating the entry and levels at the final time, it is possible to polytechnique, France exit direction, hence the angle between compute the angle between the entry and [email protected] the two, one can arbitrarily select the exit direction at the corner in order to occupation of the two energy levels. induce a transition to a state having such Ugo Boscain, CNRS, LJLL, Université Such an idea makes it possible, starting population levels. In the picture, we Pierre et Marie Curie, France from an eigenstate, to reach any super - show for instance a control inducing a [email protected] position of eigenstates whose corre - transition from an eigenstate correspon - sponding eigenvalues intersect [3]. ding to the first eigenvalue to a superpo - Mario Sigalotti sition with equal weights between the Inria, LJLL, Université Pierre et Marie Figure 2, for a 3-level system, shows a first and the third eigenstates, with no Curie, France broken adiabatic path inducing a transi - population on the second energy level. [email protected]

the Case for Quantum Software

by Harry Buhrman (CWI and QuSoft) and Floor van de Pavert (QuSoft)

Researchers and industry specialists across Europe have launched a Quantum Software Manifesto. With the Manifesto, the group aims to increase awareness of and support for quantum software research.

Quantum computers, once just the Several leading scientists and decision than 3,400 endorsements from people in dream of science fiction writers, are rap - makers, recognising the imminent prac - scientific, industrial and governmental idly becoming a reality. Already, the first tical impact of quantum technologies, organisations, the European Commission small quantum hardware devices are came together in 2016 to write the launched the Flagship Initiative on being put through their paces, with Quantum Manifesto [L1] (not to be con - Quantum Technologies. researchers probing for evidence that fused with the Quantum Software they really do work in a fundamentally Manifesto), calling for an ambitious As miraculous as the new quantum hard - different way, unlocking solutions to European quantum technology initiative ware may be, though, it can never reach problems classical computers could that would place Europe at the heart of its full potential without great quantum never solve. these new developments. After more software; this new paradigm will

34 ERCIM NEWS 112 January 2018 demand new approaches, algorithms and TheQuantumSoftware protocols. With this in mind, and spurred Manifestotcallsfor by feeling that the Flagship Initiative increasedawarenessofthe risks under-representing the crucial role importanceandurgencyof of software and theory, we have devel - quantumsoftware oped the Quantum Software Manifesto, research. which focuses on the status, outlook and specific challenges facing the quantum software field.

It is important to remember, first of all, that quantum computers are no silver bullet: where some problems will see dramatic speedups, others see none at all. Identifying potential applications and designing new quantum algorithms based on new principles will be critical, as will finding ways to achieve a useful quantum advantage on the small, noisy devices that are likely to be available in the short to medium term. We need both foundational, theoretical breakthroughs and practical work on optimising algo - developments in this area have been If any of these developments can come rithms for real quantum architectures. kicked off by the discovery of an expo - to pass, however, we will need more That, in turn, will demand ever-closer nential quantum speedup for solving good quantum programmers. Because collaboration between software and particular types of linear equations. This working with quantum computers is so hardware developers, and between aca - has led to new algorithms for core prob - fundamentally different from classical demic and industrial partners. lems such as data fitting, support vector programming, the pool of people with machines and classification. Indeed, a the necessary knowledge is currently In particular, developing reliable new quantum recommender can already small and new education programs are quantum computers is exceedingly diffi - help users of, say Netflix or Amazon, to urgently needed, in both academia and cult because their basic building blocks, find good options exponentially faster. industry. Although some initial steps called qubits, are so fragile. Even the But the real work has still has to begin in have been taken, a proposed curriculum tiniest environmental perturbation can finding applications where these tech - is still in its infancy. destroy their delicate superposition niques can be exploited. states. Software techniques for quantum We stand at the dawn of the quantum error correction and fault-tolerant com - Quantum algorithms have the potential era. It may be imminent, or may take a putation could greatly ease the task of to both break current cryptography (via little while longer to materialise, and we hardware design, helping to pave the Shor’s algorithm) and provide new and believe it is investment in quantum soft - way for large, stable quantum systems. fundamentally more secure cryptosys - ware and programmers that will make New verification and testing protocols tems, in principle even allowing users to the difference. This will help us to build based on new theoretical ideas will also run programs on untrusted systems practical hardware and develop life- be essential to both guarantee than such while keeping their data secret. Practical changing applications, but none that will devices are functioning correctly and and commercial technologies have be possible without an integrated guide their design. already been developed for quantum key approach to quantum hardware and soft - distribution and random number genera - ware development by industry and aca - Important practical applications include tion, but much work remains to fully demia. simulating physical and chemical sys - develop the potential of quantum cryp - tems, approximate optimisation and tography, particularly for large-scale The Quantum Software Manifesto can machine learning. Classical computers quantum networks. be downloaded via [L2]. find it notoriously difficult to simulate quantum systems; indeed, this currently Quantum networking has already been Link: takes up about 20% of supercomputer demonstrated, with entangled states [L1] https://kwz.me/hBj time. In contrast, little could be more being successfully distributed from [L2 ] https://kwz.me/hBq natural for a quantum computer, and this satellites to ground stations. As well as will be vital in fields such as quantum allowing truly secure communication, a Please contact: chemistry, materials science and high- global quantum internet would enable Harry Buhrman, energy physics. Many exciting experi - distributed quantum algorithms to QuSoft and CWI, The Netherlands ments are already bringing this idea exploit the power of quantum teleporta - Floor van de Pavert closer to reality. tion and error correction to solve distrib - QuSoft, The Netherlands uted tasks that require quantum +31 (0)20 592 4189 Machine learning is another hot topic in resources or for which they can make [email protected] the modern world, and a flurry of new the communication more efficient..

ERCIM NEWS 112 January 2018 35 3 6 R e s e a r c

European Research and Innovation h

a n d

I n n o v a t i o n t s i m i U p a H ( a s c a d d w s t t s I b r c a r e a h T n b a c w n d r p p s a R F n i g t s T C b C o n i h h t n h e I m o e u e i c u n n b r l a x o h n r h e r e u r e e a o y e o e e o r u e o h m e o o i i g s n e a e d e e f e n e h e c l l i q

t

d o t v n t p d l g i l d t a t m l d n g c e n s m a t e a T l o e e T o

t o

t w v o

n a f e l e i h t a i t e a

u p n s i u y e e e e e e

e o e b l

s

o s

g e

u r n d

i i i

r w m r s c r c g ) l c a d

f u s y a t t . n m f c d r t

e

l o a p i n s r a i p t e r e s s m i i t n t , d d i a

a o m e e

y o

o i i

n s r v u m s i s n t

e v o t r i

i A a

e c t g n u a s s g e n ( t n e n a y g h a s s e e w n

a d n

e

o c t

s . g

n o p l s a a e b e l w u m t t t f

a d f

e B t g a ,

s n t p g r d e n e m d r

r h v

t e m e e o n

t

u n i

e r c r t t o r l l h i A m d e a

h s r i e o t , h r . t l l i C

s i a , a h i o g e d

t a o

e o d o o

o e l m

i h s

r n

f y o n

a i s t u o a o I e o

t ( n t l a t

e o i

i t t n T a a t o p

c u m s s

m n

m

n W o n i

y

a r h s b e s y - n e g n i l h r f s s c n i t m

C a s n a t b

g i n h t t

e r v w -

h t y a

e t t

c n

f d l

e t t l m

e e

o h i e

i

a s k t y t a s

p e w i o

c o r -

l o W e i t a t a l t m e I l e i s t s

u i a i e r

s n g e c e r n a w i r p ) i f l n e

b u

s e h s o m h

o s s w r

u

t e t

n a d n a t

o r t t i o n e c a o o c o t i

a t u o e n

e o

e e t a

I e r h s m

t g p e a n b n t

i s c a t ) g i f , i o n h g e e

t r

a

a m t

r c a m e o t r l s s c f t

i b

n x t

s

a

e y

r s b o t t n t e i c o

l

g . l v s i u m n r v t d f n t a a p r i h e d

i a a f m a

t n

a n (

l m

d h

o t e o

e f i r p

o

v e

f i t o o i w e e e I n n C

n i a d r o l s t s n e e u a d d a m i o m n t t d n

h a g e

n a n f s n e o s u i r

s o e t

r g s d e h i m a e v a w l

a W r e m e

t t

t m i h a t

i

t e e i x o e e l t

t

r t o t n d u r o e i k r u e . p

a t n t y

e a a

e r p - p n r a s r S b

h e

m a a e .

w g i

f t i n

o .

n x e . g l I o i l o t ; d e r n t s a e t l o

l a f u A i c b p l a t r o s r e t ) s a r

c i s

h e

e h c o o a g s o n u y p c d l f y f d e

n

i n u r e o o c s c a h , s

l h l h s

s

i n e o s r f t p e t i

s s t g o o r o

m

e i e

q i u

t d h c h r e s

i n m c t e e i k

e s c h i e e s o s n t m a e e m c r m d

s i

n o i a t s a i

l b m o u

c

o h

l d g h s , p

m n n e d a v i i b r y g t c n g w a

a d s i u p f b

s n . o e m t b s a f i e s a t e

a f

i a t

e f g

o

g a e i b e o

. d t r e i u e p h s d s e

o f g h c e o t e t t u l a n r e

l n t l l i v e i t

s

r n c

s r e e t o p e e a

a

e T p t t t l e i u e

o o n t r n h v r l t s e

i a d e y a c f e ,

t e t w o o V u

r s r x i l d o l e u c a a

h c

e a

o t e n o

n

g t

w n

g

s . o e e

s a n t a p r u u n r e r i c h l t f

t

o e

l l d t a r g t a e i n , r s i

t o s n x b g F e o s i n t m e e e o g e h

o p s n

n h t . r i , n s e e n o n e t r r f

e o t , t i

a

t t

l s t c d t v o

e f e r

e

l w

t o a d n m i t r h i h

R

e e i o d n h n t v i p , s i l n

v a t a l b i

s d

t s a r e i

l a t

c b - a u o

n e r a e s e f m t e

f a

e u y r o i v

n

h

h t o e e

; d l t e o n r i t p i l p p e i a i n

l s u p l

h e n

r

n b n i

i c e t s

i e f i d n n c e u w

e N e e o a u m i h

f s o o r a e r t l g a e t r k g

t e

d

d

i e a

l n p e

s e

E i v r o

s h i w

s g i

o g o n t d f s o a n n t r

d b h

e r , c k o l l v t

R s N e t

a h n

o d u e l h i

p a

e y a y m

f e i o d p t a e s g a m

e

l C t m l a h i c g i s n a e

a e e h e r p

t f e i d g w e e c

r d o t e

I n

t n

s t l W

i o c e l f d c y M

v e s c r c v

a s n c

t l v e

e e

a h f e e t

y a i u e s r c s i i I s a e i i h

i d n u u p y

t o i g u o o t i e

i

r

d n i a n d d l N e n r , e e [ : n g c e t i

a a d n d d e n O E l a a e a

r

s o

r

n t n

i n n 1 l f E l h s

t

t

n e g

t , i t n p w c a d , n w

o u l n a s

a t l n u t

d

W h o

h i m t U t e

t i g s

] l b o

o c y

w

c t o

e r o

m r n n g a

o h d i g a d

n a

e S . h e f c r I e u i w s e i o i n a y e n r a d

o

n o v n

f o

d n g e l a

i v n r n s e q e i a l , A 1

u a s s

p m c j i

u n e

e a

r t a t

s e

e t i c

i n

n

e 1 c n g n o a

t i u d a

p e o p t n h

m t t t a r

c h n r o d r g

t m n n

h n p u 2 e n s I t a s . h h c

a o

i o , r o h e f

i e n

r

o o n

e v f e a o n i

n h s r

l h t a s r e y s A t c s o n

y

e

n e n t o a j t

e o c

i l o h a s o n e r c t u v J v e t i u

n p r t a

e t o

e o h e y n s

r f a l c

f b e i t t n

a u i n

f

e s f m e i t n s n o s c r e c p l e m e s e s m e n a n f u a

p o

d d c e h t d e e

t r e

t p g e u , s l t T u

c e r y t , e o n c g n

r n i w t n t s o l

w a , n i e r u o i

s t e

a y

o b r

c

a i c b e g e u

m r p m i a a g

m s e e o e s h G s e r m s r c c

o

a e s l e

r o

t o o n s y e a h e n h t o l e i d r h l n

o t i t a t a t i i c l t h

o o i u p a - r n t t s i h e d d 2 i r r r i o i o g n s i

m g n i r t a

l c o i i o c b t n i t e t

f a c e m m i

t m f l r e o 0 i n i n : n

a d a e a v a e e u n g r t h e o d o o e d o f a l s u i l a e l

1 t e i s i i n d d t e e e g d n y d d d y g s s e e e a a s s s s s s s 8 - r - - - r - - - f - - t f l l . , , adaptable energy management systems within a smart grid cooperative of homeowners. We developed an optimal and tractable decision model based on adaptive utility elicitation [2] that can find the point of diminishing returns for improving the model of user preferences. Our framework provides an extensible basis for interactive negotiation agents and is scheduled to be put in practice in 2018 within a household community in Amsterdam.

Collaborative work is currently being undertaken to extend this research further. We are organising a yearly automated negotiation competition (ANAC) [L1, L2] where uncertain preferences will act as a novel challenge for the negotiation research community. Other future work will include person - alised assistants (including for travel), autonomous driving, and making meaningful and dynamic consent workable at the Internet of Things scale.

This work is part of the Veni research programme with project number 639.021.751, which is financed by the Netherlands Organisation for Scientific Research (NWO). This work is supported by a grant from the ESPRC for Meaningful Consent in the Digital Economy project (EP/K039989/1). The research has received funding through the ERA-Net Smart Grids Plus project Grid-Friends, with support from the European Union’s Horizon 2020 research and innovation programme.

References: [1] T. Baarslag, et al.: “Computers That Negotiate on Our Figure1:Whatifwecouldnegotiateourapppermissions? Behalf: Major Challenges for Self-sufficient, Self- directed, and Interdependent Negotiating Agents”, AAMAS 2017 Workshops Visionary Papers, Lecture Notes in Computer Science. Springer International Pub - lishing, Cham, 2017. Together with The University of Southampton and The [2] T. Baarslag and M. Kaisers: “The value of information Massachusetts Institute of Technology, we are working on in automated negotiation: A decision model for eliciting new interaction mechanisms for achieving mutually benefi - user preferences”, in Proc. of the 16th Conf. on cial agreements, i.e., negotiation, as a more flexible interac - Autonomous Agents and MultiAgent Systems, tion paradigm for meaningful consent towards data sharing AAMAS’17, p. 391-400, Richland, SC, 2017. Interna - and permission management [3]. The aim is to address the tional Foundation for Autonomous Agents and Multia - inadequacy of current interaction mechanisms for handling gent Systems. data requests and obtaining consent, such as cookie notices http://dl.acm.org/citation.cfm?id=3091125.3091185 and permission pop-ups, which fail to align consumer [3] T. Baarslag, Alper T. Alan, Richard C. Gomer, Ilaria choices with their privacy preferences. The hope is to pro - Liccardi, Helia Marreiros, Enrico H. Gerding, and M.C. vide users with a more granular and iterative permission Schraefel.et al.: “Negotiation as an interaction mecha - model than current take-it-or-leave it approaches. The main nism for deciding app permissions”, in Proc. of the catalyst for improvement is a new querying model that can 2016 CHI Conference: Extended Abstracts on Human elicit preference information at the right time based on infor - Factors in Computing Systems, CHI EA’16, p. 2012- mation theoretical models from search theory. The first 2019, New York, NY, USA, 2016. ACM. results from our lab study show that users are able and http://doi.acm.org/10.1145/2851581.2892340 willing to share significantly more of their personal data when offered the benefits of negotiation, while maintaining Please contact: the same level of satisfaction with their sharing decisions. Tim Baarslag, CWI, The Netherlands Users adopt strategies in which they explore around their +31(0)20 592 4019, [email protected] desired permission set for a more nuanced negotiation that better aligns with their privacy preferences.

As a second line of research, we explore automated negotia - tions within the smart electrical grid as part of the Grid- Friends project, which is coordinated by CWI in cooperation with Fraunhofer-ITWM. The Grid-Friends team is currently developing efficient algorithms that can be used for user-

ERCIM NEWS 112 January 2018 37 Research and Innovation

in Figure 1, which can effectively map the sentence “The faster text Similarity Queen to tour Canada” to the sentence “Royal visit to Halifax”, even though (after “stopword” removal) they have Using a Linear-Complexity no words in common. WMD has been shown to work very effectively in practice, outperforming both in precision and Relaxed Word Mover’s in recall many traditional IR similarity measures. However, WMD is costly to compute because it solves a minimum cost distance flow problem, which requires cubic time in the average number of words in a document. The authors in [2] have pro - by Kubilay Atasu, Vasileios Vasileiadis, Michail Vlachos posed a lower-bound to the WMD called Relaxed WDM, or (IBM Research – Zurich) RWMD, which in practice, retrieves documents very similar to WMD, but at a much lower cost. However, RWMD still A significant portion of today’s data exists in a textual exhibits quadratic execution time complexity and can prove format: web pages, news articles, financial reports, costly when searching across millions of documents. documents, spreadsheets, etc. Searching across this collected text knowledge requires two essential For searches involving millions of documents, the RWMD components: a) A measure to quantify what is execution is inefficient because it may require repetitive considered ‘similar’, to discover documents relevant to computation of the distance across the same pairs of words. the users’ queries, b) A method for executing in real-time Pre-computing the distances across all words in the vocabu - the similarity measure across millions of documents. lary is a possibility, but it is wasteful regarding storage space. To mitigate these shortcomings, we have proposed a linear- Researchers in the Information Retrieval (IR) community complexity RWMD that avoids wasteful and repetitive com - have proposed various document similarity metrics, either at putations. Given a database of documents, where the docu - the word level (cosine similarity, Jaccard similarity) or at the ments are stored in a bag-of-words representation, the linear- character level (e.g., Levenshtein distance). A major short - complexity RWMD computes the distance between a query coming of traditional IR approaches is that they identify doc - document and all the database documents in two phases: uments with similar words, so they fail in cases when similar • In the first phase, for each word in the vocabulary, the content is expressed in a different wording. Strategies to mit - Euclidean distance to the closest word in the query docu - igate these shortcoming capitalize on synonyms or concept ment is computed based on the vector representations of ontologies, but maintaining those structures is a non-trivial the words. This phase involves a dense matrix-matrix mul - task. tiplication followed by a column- or row-wise reduction to find the minimum distances. The result is a dense vector Z. Artificial Intelligence and Deep Learning have heralded a • In the second phase, the collection of database documents new era in document similarity by capitalizing on vast is treated as a sparse matrix, which is then multiplied with amounts of data to resolve issues related to text synonymy the dense vector Z. This phase produces the RWMD dis - and polysemy. A popular approach, called ‘word embed - tances between the query document and all the database dings’, is given in [1], which maps words to a new space in documents. which semantically similar words reside in proximity to each other. To learn the embedding (i.e., location) of the words in Both phases map very well onto the linear algebra primitives the new space, those techniques train a neural network using supported by modern GPU (Graphics Processing Unit) massive amounts of publicly available data, such as devices. In addition, the overall computation can be scaled Wikipedia. Word embeddings have resolved many of the out by distributing the query documents or the reference doc - problems of synonymy. uments across several GPUs for parallel execution. Figure 2 depicts the overall approach, which enables the linear-com - In the new embedding space, each word is represented as a plexity RWMD to achieve high speeds. high-dimensional vector. To discover documents similar to a user’s multi-word query, one can use a measure called the The result is that, when computing the similarity of one doc - ‘Word-Mover’s Distance’, or WMD [2], which essentially ument with h words against n documents each having on tries to find the minimum cost to transform one text to average h words using a vector space of dimension m, the another in the embedding space. An example of this is shown complexity of the brute-force RWMD is O(nh 2m), whereas

Figure1:(Left)Mappingofword toahigh-dimensionalspaceusing wordembeddings.(Right)An illustrationoftheWordMover’s Distance.

38 ERCIM NEWS 112 January 2018

DatabaseDatat)abase documentsdocuments (D)(D Figure2: Overviewofourapproach.

PhasePhase 1 PhasePhase 2 !!"#$%&'($)"#*$+,$"#$$$$%&'( )"#* +, 95%;4&-#+<$96%5;&# 4&-#+<$ --(%$."'&/01$(%$$."'&/01 *%,6%;.%'-"#$*;.%%,6% .%'-"#$ QueryQyuentr ddocumentsocument)s ((Q)Q ''"450-%$-(%$"450-%$$-(% 401-+51+'&-+",$401-+51+''&&-+",$ **+6-&,'%$-"$-(%$+6-&,'%$$-" -(% $ /%-)%%,$=/%-)%%,$=%%,$ &,*$ ''1"6%6-$)"#*$+,$1"6%6-$$$)"#* +, DenseDense :$ -"$'"450-%$-"$$'" 450-% vectorvector Z ",%7 70%#20%# 2 * *"'8$" '8$ -(%$*+6-&,'%6$-(%$*+6-& ,'%6$ 99-"#%$-(%$#%601-6$-"#%$$-(% #%601-6$ /%-)%%,$-(%$/%-)%%,$-(%$%%,$ ++,$&$*%,6%$,$$& *%,6% $ 770%#2$&,*$-(%$0%#2$&,*$$-(% ..%'-"#$:8%'-"#$:8 **&-&/&6%$*"'68&-&/&66%%$*"'68 s s U U P P G G

f f o o

. . . r r e e t DistributeDistribuQte Q t s s u u l acrossasscro GPUsGPUs l C C

under our methodology the complex ity is O(nhm ). The inter - The results suggest that we could use the high-quality ested re ader can find a dditional tech nical de tails about the matches of the RW MD to q uer y – in sub- second time – at linear complexity RWMD implementation in [3]. least 100 mi llion document s usi ng only a modest com pu ta -

tional in fras truc ture. To sh owcase th e p erformance of the new m ethod olo gy , we have used very l arg e datasets of news docum ents. On e doc u - Refe ren ces: ment is pose d as t he query and the search r etrieves t he k- [ 1 ] T. M ikolo v, K. Chen , G. Co rra do, a nd J. Dean: “E ffi - most-similar documents in the database. Such a scenario can cient estimation of word representations in vector

be used either for performing duplicate dete ction (whe n the spac e”, CoRR, ab s/130 1.3 781, 2013. distance is bel ow a th reshold), or f or a chieving clu sterin g of [2] M. J . Kusner, et al.: “From wor d embeddings to do cu - ! similar news ev ents. We conducte d our e xp eriments on a men t distanc es, I CML 2015: 957 –966. cluster of 16 NVIDIA Tesla P100 GPUs on two dataset s. S et [3 ] K. Atasu, e t al .: “Linear-Co mplexit y R elaxe d Word

1 comprises 1 mil lion documents, w ith a n average n umb er o f Mover ’s Distan ce with GPU Acc eleration”, IE EE B ig 108 words per document. Set 2 comprises 2.8 million words , D at a 201 7. with an ave ra ge numbe r of 28 words p er •d!o cum ent. The comparison of th e runtim e perform ance bet ween WM D , Pl eas e co ntact : R WM D, and our solution is given in Figure 3, which sh ows Kub il ay At asu, I BM R ese arc h – Z ur ic h, Sw itze rla nd that th e prop osed linear-comp lexity RWMD can be more kat@ zu rich.ibm .com than 70 times faste r than the orig inal R WMD and more th an 16,000 times faster than the Word Mover’s Distance.

)*+#,-"$./-012,3/+

%''

%' &=>?

< A@?

3 % 456 ; "

3 )456 : $

" '(% 7.8)456 - , @%? &&? 9

'('% Figure3:Timetocomputethek-most- similardocumentsusing16NVIDIATesla ' P100GPUs. !"#$% !"#$&

ERCIM NEWS 112 January 2018 39

[

Research and Innovation

the CAPtCHA Samples Website

by Alessia Amelio (University of Calabria), Darko Brodić, Sanja Petrovska (University of Belgrade), Radmila Janković (Serbian Academy of Sciences and Arts)

“CAPTCHA Samples” is a new website for testing different types of CAPTCHA specifically designed for research and study purposes. Figure2:WebpageofCAPTCHASampleswithamessagereporting successinsolvingthetest,includingcompletiontimeandnumberof In recent decades, the CAPTCHA test has received much attempts. attention owing to the increasing security problems affecting the web, for which risk prevention plays a key role. tries to find the correct solution. Figure 2 shows an example of CAPTCHA is an acronym of “Completely Automated Public a successfully solved “surprised face” CAPTCHA test, which Turing Test to tell Computers and Humans Apart”. This is a was solved in 2.48 seconds and one attempt. test program which an internet user is required to solve in order to prove that he or she is human and thus gain access to Every test is completely anonymous because any personal a given website. In fact, the CAPTCHA test is designed to be information is registered about the internet user who easily solved by a human and difficult to solve for an auto - accesses the CAPTCHA. Also, the main advantage of the matic program which tries to obtain unauthorized access to website is that the CAPTCHA tests can be accessed from dif - the website [1]. Recently, different types of CAPTCHA have ferent devices, including smartphones, laptops and tablets. In flourished in the literature, with more sophisticated security this way, the difference in terms of CAPTCHA usability mechanisms that guarantee efficacy in risk prevention, and between multiple devices can be easily checked. In the more user-friendly interfaces [2], [3]. In order for researchers future, we are planning to extend the CAPTCHA Samples to effectively build on the recent work in this area, an analysis website with new functionalities, including the selection of of the most recently introduced CAPTCHA tests is needed. other CAPTCHA tests of different typology, which will be added to the website. It will be particularly useful in aca - Accordingly, we present CAPTCHA Samples, a new website demic as well as industrial research for anonymously gath - collecting different CAPTCHA tests for research and ering data about the usability of CAPTCHA tests. It could analysis purposes. CAPTCHA Samples, available at [L1] also be the starting point for designing new CAPTCHA types (see Figure 1), was born as a joint project involving the which are appropriate for specific types of internet users. For Technical Faculty in Bor, University of Belgrade, Serbia, and more detailed information about the image-based DIMES University of Calabria, Italy. CAPTCHA tests included in the CAPTCHA Samples web - site, please refer to [3]. Currently, the website reports different samples of image- based CAPTCHA tests, including tests based on facial Link: expressions (animated characters, face of an old woman, sur - [L1] http://captchasamples.altervista.org prised face, and worried face). The aim of these CAPTCHA tests is to recognise the correct image from a list of different References: images, according to the question asked by the test. From the [1] A.M. Turing: “Computing Machinery and Intelligence”, homepage, it is possible to select the CAPTCHA test of Mind 59:433-460, 1950. interest and access to the corresponding webpage. The [2] D. Brodić, S. Petrovska, M. Jevtić, Z. N. Milivojević: CAPTCHA test can be solved in order to obtain useful infor - “The influence of the CAPTCHA types to its solving mation about the usability of the test, including: (i) comple - times”, MIPRO: 1274-1277, 2016. tion time, which is the solution time to the CAPTCHA, and [3] D. Brodić, A. Amelio, R. Janković: “Exploring the (ii) number of tries to provide the correct solution to the Influence of CAPTCHA Types to the Users Response CAPTCHA. Time by Statistical Analysis”, Multimedia Tools and Applications, 1-37, 2017. If the CAPTCHA test is not correctly solved, then the failure is notified by a text field of “wrong answer” on the screen, and Please contact: the user is asked to try again. Otherwise, if the CAPTCHA test Alessia Amelio, University of Calabria, Rende, Italy is correctly solved, a message in a text field on the screen is [email protected] visualised, reporting the completion time and the number of

Figure1:CAPTCHA Sampleswebsite.

40 ERCIM NEWS 112 January 2018 approach that has its roots in the IBIS-style argumentation APoPSIS: A Web-based model [1], but with slightly different semantics, is designed to structure the argumentation process [2] and organise the Platform for the Analysis different conceptual components of the dialogue. In order to be able to accommodate (store and retrieve) all these compli - of Structured dialogues cated opinions, a formal semantic web ontology, called MACE (Multi Aspect Comment Evaluation), was designed, by Elisjana Ymeralli, Theodore Patkos and Yannis which can semantically represent the content produced and Roussakis (ICS-FORTH) exchanged within the platform for online communities. Moreover, the system uses the s-mDiCE (symmetric multi- APOPSIS is a web-based platform that aims to motivate Dimensional Comment Evaluation) argumentation frame - online users to participate in well-structured dialogues by work [3], for evaluating online discussions and identifying raising issues and posting ideas or comments, related to the strongest and most acceptable opinions found in online goal-oriented topics of discussion. The primary goal of the debates, based on different metrics. The underlying quantita - system is to offer automated opinion analysis features tive algorithms are generic enough to capture the features of that help identify useful patterns of relations amongst online communities on the social web and may benefit sev - participants and their expressed opinions. Our system is eral platforms, from debate portals to decision-making sys - designed to enable more structured and less confusing tems. argumentative discussions, thus helping sense-makers in understanding the dynamic flow of the dialogue. A key feature of APOPSIS is the support for automated opinion analysis for analysing user behaviour in online com - Web-based platform APOPSIS can be used in everyday munities. Opinion analysis aims at identifying different deliberation, as well as decision-making discussions, where groups of users and useful patterns of relationship between users exchange their viewpoints and argue over a plethora of participants and opinions, in order to help users to make topics. The platform offers well-structured dialogues which better sense of the dialogue and the opinion exchange can take place in different phases of discussions. As a process. Based on user reactions, the system applies ML debating platform, APOPSIS offers the opportunity for a variety of groups to work and collaborate with each other by sharing their ideas in support of or against other opinions. Each dialogue is open for users to defend and justify their state - ments and vote upon them, respectively. The first phase of the dialogue allows users to participate by providing solu - tions or statements, in addition to their justified agreements or disagreements, while in the second phase, only the most popular or well-justified solutions remain and become subject to more detailed evaluation by the participants.

Conversations are presented in a tree-like style, where subsequent levels of com - ments respond to the parent comment. Figure1:ApartofCityPlanningdialogue. The system supports several features that enrich the system’s usability, such as voting, semantic annotation, aspect-based rating and searching functionalities. These features enable APOPSIS to pro - duce and extract useful conclusions that can help sense-makers and expert users to understand and make decisions on spe - cific problems and issues.

In order to interpret and analyse user reactions, such as comments, votes and ratings, the system uses methods from the fields of computational argumentation and machine learning (ML) that enable the structuring and evaluation of the dif - fering opinions expressed in dialogues. Specifically, an argumentation-based Figure2:AnautomatedOpinionsAnalysis.

ERCIM NEWS 112 January 2018 41 Research and Innovation

algorithms for the clustering of features and the extraction of association rules. Can we trust Machine The basic ML algorithms used in this platform are: Learning Results? • Expectation-Maximisation algorithm (EM): Decides the optimal number of clusters. Artificial Intelligence • K-means algorithm: Identifies different groups of users and opinions. in Safety-Critical decision • Apriori algorithm: Determines interesting and useful rela - tionships among attributes. Support

Our methodology implements five different information by Katharina Holzinger (SBA Research), Klaus Mak, needs emerging from users throughout the sense-making (Austrian Army) Peter Kieseberg (St. Pölten University of process. Applied Sciences), Andreas Holzinger (Medical University • Same profiles with similar opinions: Identifies different Graz, Austria) user profiles that share similar opinions on specific posi - tions. Machine learning has yielded impressive results over the • Sharing similar opinions with specific user: Extracts dif - last decade, but one important question that remains to ferent groups of users whose opinions are closely related be answered is: How can we explain these processes and to specific user(s). algorithms in order to make the results applicable as • Same users with similar preferences: Determines groups proof in court? of users who share both similar opinions and profile char - acteristics. Artificial intelligence (AI) and machine learning (ML) are • Similar opinions based on different profiles: Identifies making impressive impacts in a range of areas, such as similar opinions expressed by users with different profile speech recognition, recommender systems, and self-driving characteristics. cars. Amazingly, recent deep learning algorithms, trained on • Different user profiles with similar/dissimilar opinions: extremely large data sets, have even exceeded human per - Finds different groups of user profiles who share similar formance in visual tasks, particularly on playing games such or dissimilar opinions. as Atari Space Invaders, or mastering the game of Go [L1]. An impressive example from the medical domain is the There are several avenues that are worthy of further investi - recent work by Esteva et al. (2017) [1]: they utilised a gation, with an emphasis on improving the system’s GoogleNet Inception v3 convolutional neural network usability. Thorough evaluations with real users and large (CNN) architecture for the classification of skin lesions, pre- datasets of discussions will be carried out, including expert trained their network with approximately 1.3 million images walk-through evaluation of the platform and adjustments (1,000 object categories), and trained it on nearly 130,000 based on user expert feedback. clinical images. The performance was tested against 21 board-certified dermatologists on biopsy-proven clinical Links: images. The results show that deep learning can achieve a [L1] www.ics.forth.gr/isl/apopsis performance on par with human experts. [L2] http://www.ics.forth.gr/isl/mace/mace.rdfs [L3] http://www.ics.forth.gr/isl/mace/MACE%20Ontol - One problem with such deep learning models is that they are ogy_Scope_Notes.pdf considered to be “black-boxes”, lacking explicit declarative knowledge representation, hence they have difficulty in gen - References: erating the required underlying explanatory structures. This [1] J. Conklin, M.L. Begeman: “gIBIS: A hypertext tool for is limiting the achievement of their full potential, and even if team design deliberation”, Proc. of the ACM confer - we understand the mathematical theories behind the machine ence on Hypertext, ACM, 1987. model, it is still complicated to get insight into the internal [2] J.Schneider, T.Groza, and A.Passant: “A review of argu - workings. Black box models lack transparency and one ques - mentation for the social semantic web”, Semantic Web tion is becoming increasingly important: “Can we trust the 4.2 (2013) results?” We argue that this question needs to be rephrased [3] T.Patkos, G.Flouris, and A.Bikakis: “Symmetric Multi- into: “Can we explain how and why a result was achieved?” Aspect Evaluation of Comments”, ECAI 2016-22nd (see Figure 1). A classic example is the question “Which European Conference on Artificial Intelligence, 2016. objects are similar?” This question is the typical pretext for using classifiers to classify data objects, e.g. pictures, into Please contact: different categories (e.g., people or weapons), based on util - Elisjana Ymeralli, FORTH, Greece ising implicit models derived through training data. Still, an [email protected] (mailto:[email protected]) even more interesting question, both from theoretical and legal points of view, is “Why are those objects similar?” This is especially important in situations when the results of AI are not easy to verify. While verification of single items is easy enough in many classical problems solved with machine learning, e.g., detection of weapons in pictures, it can be a problem in cases where the verification cannot be

42 ERCIM NEWS 112 January 2018 nals of the intrinsic model built by these algorithms. Furthermore, and in spirit with our second research interest in this area, the deletion of data is a highly complicated process in most modern complex environ - ments and gets even more complicated when considering the typical targets of data provisioning environments like data - bases that are opposing deletion: • Fast searches: Typically, one main goal in database design is to provide fast and efficient data retrieval. Thus, the internal structures of such systems have been designed in order to speed up the search process by incorporating parts of the content as search keys, Figure1:AIshowsimpressivesuccess,stillhavingdifficultyingeneratingunderlying yielding a tree structure that is organ - explanatorystructures. ised along the distribution of key infor - mation, thus making deletion a prob - lematic issue. done by a human with (close to) 100 percent precision, as in • Fast data manipulation: Like in modern file systems, data cancer detection, for example. entries that are “deleted” are not actually erased from the disk with overwriting the respective memory for perform - Consequently, there is growing demand for AI, which not ance reasons, but only unlinked from the search indices only performs well, but is also transparent, interpretable and and marked for overwriting. trustworthy. In our recent research, we have been working on • Crash Recovery: Databases must possess mechanisms in methods and models to reenact the machine decision-making case an operation fails (e.g., due to lack of disk space) or process [2], to reproduce and to comprehend the learning and the database crashes in the middle of a data-altering oper - knowledge extraction processes, because for decision sup - ation (e.g., blackouts) by reverting back to a consistent port it is very important to understand the causality of state that is throughout all data tables. In order to provide learned representations. If human intelligence is comple - this feature, transaction mechanisms must store the data mented by machine learning, and in some cases even over - already written to the database in a crash-safe manner, ruled, humans must be able to understand, and most of all to which can be used, for example, in forensic investigations be able to interactively influence the machine decision to uncover deleted information. process. This needs sense making to close the gap between • Data Replication: Companies have implemented mirror human thinking and machine “thinking”. data centres that contain replicated versions of the opera - tive database in order to be safe against failures. Deletion This is especially important when the algorithms are used to from such systems is thus especially complicated. extract evidence from data to be used in court. A similar dis - cussion has already been started in the area of digital foren - With our research, we will be able to generate mechanisms sics, with novel forensic processes being able to extract small for better understanding and control of the internal models of parts of information from all over the IT-system in question machine learning algorithms, allowing us to apply fine- and then combining them in order to generate evidence – grained changes on one hand, and to better estimate the which is currently not usable in court, simply owing to the impact of changes in knowledge bases on the other hand. In fact that no judge will rule on evidence gathered by a process addition, our research will yield methods for enforcing the no one can control and see through. right to be forgotten in complex data driven environments.

Our approach will also address rising legal and privacy con - Link: [L1] https://arxiv.org/abs/1708.01104 cerns, e.g., with the new European General Data Protection Regulation (GDPR and ISO/IEC 27001) entering into force on References: May, 25, 2018. This regulation will make black-box [1] A. Esteva, et al.: “Dermatologist-level classification of approaches difficult to use in business, because they are not skin cancer with deep neural networks”, Nature, 542, able to explain why a decision has been made. In addition, it (7639), 115-118, 2017, doi:10.1038/nature21056. must be noted that the “right to be forgotten” [3] established by [2] A. Holzinger, et al.: “A glass-box interactive machine the European Court of Justice has been extended to become a learning approach for solving NP-hard problems with “right of erasure”; it will no longer be sufficient to remove a the human-in-the-loop”, arXiv:1708.01104. person’s data from search results when requested to do so, data [3] B. Malle, P. Kieseberg, S. Schrittwieser, A. Holzinger: controllers must now erase that data. This is especially prob - “Privacy Aware Machine Learning and the ‘Right to be lematic when data is used inside the knowledge base of a Forgotten’”, ERCIM News 107, (3), 22-23, 2016. machine learning algorithm, as changes here might make deci - sions taken and results calculated by the algorithm irrepro - Please contact: ducible. Thus, in order to understand the impact of changes to Katharina Holzinger, SBA Research, Vienna, Austria such results, it is of vital importance to understand the inter - [email protected]

ERCIM NEWS 112 January 2018 43 Research and Innovation

the longer the braking distance and the longer the blocks formal Methods need to be, thus decreasing the line’s capacity. This is because stringent safety requirements impose the length of for the Railway Sector fixed blocks to be based on the worst-case braking distance, regardless of the actual speed of the train. by Maurice ter Beek, Alessandro Fantechi, Alessio Ferrari, Stefania Gnesi (ISTI-CNR, Italy), and Riccardo With a moving block signalling system, in contrast, a safe Scopigno (ISMB, Italy) zone around the moving train can be computed, thus opti - mising the line’s exploitation (Figure 1). For this solution to Researchers from the Formal Methods and Tools group work, it requires the precise absolute location, speed and of ISTI-CNR are working on a review and assessment of direction of each train, to be determined by a combination of the main formal modelling and verification languages sensors: active and passive markers along the track, as well and tools used in the railway domain, with the aim of as trainborne speedometers. One of the current challenges in evaluating the actual applicability of the most promising the railway sector is to make moving block signalling sys - ones to a moving block signalling system model provided tems as effective and precise as possible, including GNSS by an industrial partner. The research is being conducted and leveraging on an integrated solution for signal outages in the context of the H2020 Shift2Rail project ASTRail. (think, e.g., of tunnels) and the problem of multipaths [1]. This is one of the main topics addressed by the project Compared with other transport sectors, the railway sector is ASTRail: SAtellite-based Signalling and Automation notoriously cautious about adopting technological innova - SysTems on Railways along with Formal Method and tions. This is commonly attributed to the sector’s robust Moving Block Validation (Figure 2). A subsequent aim of the safety requirements. An example is smart route planning: project is to study the possibility of deploying the resulting while GNSS-based positioning systems have been in use for precise and reliable train localisation to improve automatic quite some time now in the avionics and automotive sectors driving technologies in the railway sector. to provide accurate positioning and smart route planning, the current train separation system is still based on fixed blocks – We are currently reviewing and assessing the main formal a block being the section of the track between two fixed modelling and verification languages and tools used in the points. In signalling systems, blocks start and end at signals, railway domain in order to identify the ones that are most with their lengths designed to allow trains to operate as fre - mature for application in the railway industry [2] [3]. This is quently as necessary (i.e., ranging from many kilometres for done by combining a scientific literature review with inter - secondary tracks to a few hundred metres for a busy com - views with stakeholders of the railway sector. This will muter line). The block sizes are determined based on param - shortly result in a classification and ranking, from which we eters such as the line’s speed limit, the train’s speed, the will select a set of languages and tools to be adopted in all the braking characteristics of trains, sighting and reaction time relevant development stages of the systems addressed by of drivers, etc. However, the faster trains are allowed to run, ASTRail. In particular, we will formalise and validate a

Figure1:Safebrakingdistancebetween trainsinfixedblockandmovingblock signallingsystems(Imagecourtesyof Israel.abad/WikimediaCommons distributedundertheCCBY-SA3.0 license).

Figure2:Illustrativesummary ofASTRail’sobjectives.

44 ERCIM NEWS 112 January 2018 moving block model developed by SIRTI, which is an industry leader in the design, production, installation and the Impacts of Low- maintenance of railway signalling systems. Quality training data ASTRail will run until August 2019 and is coordinated by Riccardo Scopigno from the Istituto Superiore Mario Boella on Information Extraction sulle Tecnologie dell’Informazione e delle Telecomuni- cazioni (ISMB, Italy). Other partners are SIRTI S.p.A. from Clinical Reports (Italy), Ardanuy Ingeniería S.A. (Spain), École Nationale de l’Aviation Civile (ENAC, France), Union des Industries by Diego Marcheggiani (University of Amsterdam) and Ferroviaires Européennes (UNIFE, Belgium) and ISTI-CNR Fabrizio Sebastiani (CNR) (Italy). In a joint effort between the University of Amsterdam Link: and ISTI-CNR, researchers have studied the negative ASTRail: http://www.astrail.eu/ impact that low-quality training data (i.e., training data annotated by non-authoritative assessors) has on References: information extraction (IE) accuracy. [1] F. Rispoli, et al.: “Recent Progress in Application of GNSS and Advanced Communications for Railway Sig - Information Extraction (IE) is the task of designing software naling”, in 23rd Intl. Conf. Radioelektronika, IEEE, artifacts capable of extracting, from informal and unstructured 2013, 13-22. DOI: 10.1109/RadioElek.2013.6530882 texts, mentions of particular concepts, such as the names of [2] A. Fantechi, W. Fokkink, and A. Morzenti: “Some people, organisations and locations – where the task usually Trends in Formal Methods Applications to Railway Sig - goes by the name of “named entity extraction”. Domain-spe - naling”, Chapter 4 in Formal Methods for Industrial cific concepts, such as drug names, or drug dosages, or com - Critical Systems: A Survey of Applications (S. Gnesi plex descriptions of prognoses, are also examples. and T. Margaria, eds.). John Wiley & Sons, 2013, 61- 84. DOI: 10.1002/9781118459898.ch4 Many IE systems are based on supervised learning, i.e., rely [3] A. Fantechi: “Twenty-Five Years of Formal Methods on training an information extractor with texts where men - and Railways: What Next?”, in Software Engineering tions of the concepts of interest have been manually and Formal Methods, LNCS, vol. 8368. Springer, 2013, “labelled” (i.e., annotated via a markup language); in other 167-183. DOI: 10.1007/978-3-319-05032-4_13 words, the IE system learns to identify mentions of concepts by analysing what manually identified mentions of the same Please contact: concepts look like. Stefania Gnesi, ISTI-CNR, Italy [email protected] It seems intuitive that the quality of the manually assigned labels (i.e., whether the manually identified portions of text are indeed mentions of the concept of interest, and whether their starting points and ending points have been identified precisely) has a direct impact on the accuracy of the extractor that results from the training. In other words, one would expect that learning from inaccurate manual labels will lead

Figure1: Screenshot displayinga radiological reportannotated accordingtothe conceptsof interest.

ERCIM NEWS 112 January 2018 45 Research and Innovation

to inaccurate automatic extraction, according to the familiar 4th International Indoor “garbage in, garbage out” principle. Positioning and Indoor However, the real extent to which inaccurately labelled training data impact on the accuracy of the resulting IE Navigation Competition system, has seldom (if ever) been tested. Knowing how much accuracy we are going to lose by deploying low-quality labels The fourth international Indoor Positioning and Indoor is important, because low-quality labels are a reality in many Navigation (IPIN) competition, seventh in the EvAAL real-world situations. Labels may be low quality, for example, series, was hosted by the international IPIN conference when the manual labelling has been performed by “turkers” in Sapporo, Japan, on 17 September 2017. (i.e., annotators recruited via Mechanical Turk or other crowdsourcing platforms), or by junior staff or interns, or The aim of the competition is to measure the performance of when it is old and outdated (so that the training data are no indoor localisation systems, that are usable in offices, hospi - longer representative of the data that the information extractor tals or other big buildings like warehouses. The 2017 edition will receive as input). What these situations share in common has attracted 28 teams and allowed participants to test their is that the training data were manually labelled by one or localisation solutions with rigorous procedures inside the more “non-authoritative” annotators, i.e., by someone dif - two-floor structure of the Conference Hall of Hokkaido ferent from the (“authoritative”) person who, in an ideal situa - University. tion (i.e., one where there are no time / cost / availability con - straints), would have annotated it.

In this work, the authors perform a systematic study of the extent to which low-quality training data negatively impacts on the accuracy of the resulting information extractors. The study is carried out by testing how accuracy deteriorates when training and test set have been annotated by two dif - ferent assessors. Naturally enough, the assessor who anno - tates the test data is taken to be the “authoritative” annotator (since accuracy is tested according to her/his judgment), while the one who annotates the training data is taken to be the “non-authoritative” one. The study is carried out by IPINparticipants. applying widely used “sequence learning” algorithms (either Conditional Random Fields or Hidden Markov Support Vector Machines) on a doubly annotated dataset (i.e., a The competition ended with the awarding of four 150.000¥ dataset in which each document has independently been (1.100€) prizes: annotated by the same two assessors) of radiological reports. • smartphone-based (Chan Gook Park, Seoul National Uni - Such reports, and clinical reports in general, are generated by versity, Corea) clinicians during everyday practice, and are thus a chal - • dead reckoning (Chuanhua Lu, Kyushu University, Japan) lenging type of text, since they tend to be formulated in • offline smartphone-based (Adriano Moreira, University of informal language and are usually fraught with typos, idio - Minho, Portugal) syncratic abbreviations, and other types of deviations from • offline PDR warehouse picking (Yoshihiro Ito, KDDI linguistic orthodoxy. The fact that the dataset is doubly anno - R&D Laboratories Inc., Japan). tated allows the systematic comparison between high-quality settings (training set and test set annotated by annotator A) Prizes were awarded by the official sponsors of the competi - and low-quality settings (training set annotated by annotator tion KICS , ETRI, TOPCON, and PDR Benchmark, respec - NA and test set annotated by annotator A), thereby allowing tively. precise quantification of the difference in accuracy between the two settings. Francesco Potortì, Antonino Crivello and Filippo Palumbo, researchers at the WNLab group of CNR-ISTI at Pisa, were Link: among the main organisers. http://nmis.isti.cnr.it/sebastiani/Publications/JDIQ2017.pdf More information on the international competition, includ - References: ing complete results, photos and comments at [1] A. Esuli, D. Marcheggiani, F. Sebastiani: “An http://evaal.aaloa.org Enhanced CRFs-based System for Information Extrac - tion from Radiology Reports”, Journal of Biomedical Informatics 46, 3 (2013), 425–435. [2] W. Webber, J. Pickens: “Assessor disagreement and text classifier accuracy”, in Proc. of SIGIR 2013,929–932, 2013.

Please contact: Fabrizio Sebastiani, ISTI-CNR, Italy +39 050 6212892 [email protected]

46 ERCIM NEWS 112 January 2018 Events

Joint 22nd International Workshop on formal Methods for Industrial Critical Systems and 17th International Workshop on Automated verification of Critical Systems by Ana Cavalcanti (University of York), Laure Petrucci (LIPN, CNRS & Université Paris 13) and Cristina BestpaperawardwinnersEtienneAndré(left)andTimWillemse. Seceleanu (Mälardalen University)

The yearly workshop of the ERCIM Working Group on Wesselink, Rob Wieringa and Tim Willemse. An excellent Formal Methods for Industrial Critical Systems (FMICS) programme together with a nice weather contributed to the was organised as a joint event together with the success of the workshops. workshop on Automated Verification of Critical Systems (AVoCS). The resulting FMICS-AVoCS 2017 workshop took We gratefully acknowledge the support of Springer for pub - place on 18-20 September in Turin, hosted by the lishing the workshop’s proceedings and sponsoring the best University of Turin. papers, EasyChair for assisting us in managing the complete process from submission to proceedings, as well as ERCIM The aim of the FMICS workshop series is to provide a forum and EASST. for researchers interested in the development and application of formal methods in industry. It strives to promote research The proceedings of FMICS-AVoCS 2017 have been pub - and development for the improvement of formal methods lished by Springer as volume 10471 of their LNCS series. and tools for industrial applications. The aim of the AVoCS workshop series is to contribute to the interaction and Selected papers are proposed for special issues of the inter - exchange of ideas among members of the international com - national journals Software Tools for Technology Transfer munity on tools and techniques for the verification of critical (STTT) and Science of Computer Programming (SCP). systems. Links: The workshop was chaired by Laure Petrucci (LIPN, CNRS FMICS-AVoCS 2017: & Université Paris 13, France) and Cristina Seceleanu http://www.es.mdh.se/conferences/fmics-avocs-2017/ (Mälardalen University, Sweden). A special track on “Formal FMICS Working Group: http://fmics.inria.fr/ methods for mobile and autonomous robots” took place within the event, and was chaired by Ana Cavalcanti Reference: (University of York, UK). The workshop attracted 30 partici - Laure Petrucci, Cristina Seceleanu and Ana Cavalcanti pants from eleven countries. (eds.): “Critical Systems: Formal Methods and Automated Verification - A total of thirty papers were submitted, including eight Joint 22nd International Workshop on Formal Methods for specifically for the special track. Fourteen of them were Industrial Critical Systems and 17th International Work - accepted (including four for the special track). shop on Automated Verification of Critical Systems (FMICS-AVoCS’17)”, Turin, Italy, 18–20 September 2017, The programme also included two excellent invited keynote Lecture notes in Computer Science 10471, Springer, 2017. lectures: “Replacing store buffers by load buffers in total https://doi.org/10.1007/978-3-319-67113-0 store ordering” by Parosh Abdulla (Uppsala University, Sweden) and “Towards formal apps: turning formal methods Please contact: into verification techniques that make the difference in prac - Ana Cavalcanti, University of York tice” by Kerstin Eder (University of Bristol, UK). Moreover, [email protected] a half-day tutorial took place: “DIME: Model-based genera - tion of running web applications” by Tiziana Margaria Laure Petrucci, LIPN, CNRS & Université Paris 13 (University of Limerick & Lero, Ireland) and Philip [email protected] Zweihoff (TU Dortmund, Germany). Cristina Seceleanu, Mälardalen University The presentations were of extremely good quality. The pro - [email protected] gramme committee awarded two best papers: “A unified for - malism for monoprocessor schedulability analysis under uncertainty” by Etienne André, and “Formalising the Dezyne modelling language in mCRL2” by Rutger van Beusekom, Jan Friso Groote, Paul Hoogendijk, Rob Howe, Wieger

ERCIM NEWS 112 January 2018 47 Events

10th International visual Heritage 2018 IfIP Networking 2018

Conference of the Vienna, 12-15 November 2018 The IFIP Networking 2018 Conference (NETWORKING 2018), to be held in ERCIM Working Group With the aim of continuing the suc - Zurich, Switzerland, from 14-16 May cessful experience of Digital Heritage 2018 is the 17th event of the series, spon - on Computational 2013 (Marseille, France) and Digital sored by the IFIP Technical Committee Heritage 2015 (Granada, Spain), the on Communication Systems (TC6). and Methodological next edition of the Eurographics Accepted papers will be published in the Graphics and Cultural Heritage (EG IFIP Digital Library and submitted to the Statistics GCH) Symposium will be organized in IEEE Xplore Digital Library. cooperation with CHNT (Cultural The 10th International Conference of the Heritage and New Technologies) in The main objective of Networking 2018 ERCIM WG on Computational and Vienna. The aim of this federated event is to bring together members of the net - Methodological Statistics (CMStatistics is again to bring different communities working community, from both aca - 2017) took place at the Senate House in the same venue, to share experiences demia and industry, to discuss recent and Birkbeck, University of London, and discuss methodologies concerning advances in the broad and quickly- UK, 16-18 December 2017. Tutorials digital visual media and their use in the evolving fields of computer and commu - were given on Friday 15th of December context of cultural heritage applica - nication networks, to highlight key 2017 and the COST IC1408 CRoNoS tions. issues, identify trends, and develop a Winter Course on Copula-based model - vision for future Internet technology, ling with R took place the 13-14 We are looking for a wide participation operation, and use. December 2017. The conference took of our community, as well as the collab - place jointly with the 11th International oration and inclusion of other structured Important Dates: Conference on Computational and communities and events, since Visual • Abstract registration: January 2, 2018 Financial Econometrics (CFE 2017). Heritage 2018 is aimed as an open circle (everywhere on earth) (please contact the organizers at visual - • Full paper submission: This annual conference has become a [email protected] if you are an January 7, 2018 leading joint international meeting at the organization/community interested in • Acceptance notification:March 5, 2018 interface of statistics, econometrics, joining us). • Camera-ready papers: March 23, 2018 empirical finance and computing. The conference aims at bringing together The 2018 edition will be a special event, Link: researchers and practitioners to discuss since 2018 has been declared by the http://networking.ifip.org/2018/ recent developments in computational European Commission the "European methods for economics, finance, and Year of Cultural Heritage". The event in statistics. The CFE-CMStatistics2017 Vienna will also take place during the programme comprised of 375 sessions, Austrian term as President of the EC. Web Science 5 plenary talks and over 1550 presenta - Therefore, VH2018 will be an ideal tions. There were about 1700 partici - context for discussing European poli - Conference 2018 pants. This was the biggest meeting of cies on digital heritage and digital the conference series in terms of number humanities. Amsterdam, 27-31 May 2018 of participants and presentations. The growth of the conference in terms of size Visual Heritage 2018 will have inde - The 10th International ACM Conference and quality makes it undoubtedly one of pendent call of papers for each event on Web Science in 2018 (WebSci’18) is the most important international scien - contributing to the program. The EG a unique conference where a multitude tific events in the field. GCH 2018 program will be based on a of disciplines converge in a creative and call for papers (full and short papers) critical dialogue with the aim of under - Link: with deadline June 20th, 2018 standing the Web and its impacts. The http://www.cmstatistics.org/ conference brings together researchers Paper submission from multiple disciplines, like computer Please contact: More details on the EG GCH scientific science, sociology, economics, informa - [email protected] topics, instructions for submitters, call tion science, anthropology and psy - for paper dates, and the selection proce - chology. Web Science is the emergent dure will be published soon on EG GCH study of the people and technologies, VH2018 web page http://2018.visual - applications, processes and practices heritage.org that shape and are shaped by the World Wide Web. More information http://www.chnt.at/ Keynote speakers include: Tim Berners- Lee, José van Dijck and John Domingue

More information: https://websci18.webscience.org

48 ERCIM NEWS 112 January 2018 2018 ICOM CIDOC international committee for Conference documentation Postdoc Position in the Research Project CIdoC Conference “Approximation Algorithms, Quantum Information 2018 and Semidefinite optimization”

Heraklion, Crete, Greece, 29 Centrum Wiskunde & Informatica (CWI) has a vacancy in the Network & Optimization research September – 4 October 2018 group for a talented Postdoc in the research project "Approximation Algorithms, Quantum Information and Semidefinite Optimization”. Conference theme: Provenance of Knowledge Job description The research project “Approximation algorithms, quantum information and semidefinite opti - Participants in the event are invited to mization” aims to explore the limits of efficient computation within classical and quantum com - consider the theme of this conference, puting, using semidefinite optimization as a main unifying tool. The position involves research ‘Provenance of Knowledge’, both in its into the mathematical and computer science aspects of approximation algorithms for discrete broad sense and in its technical detail. optimization, quantum entanglement in communication, and complexity of fundamental prob - Today, museum professionals have to lems in classical and quantum computing. The research will be supervised by Prof. Monique take into account as much the traditional Laurent from the CWI Networks & Optimization research group, in collaboration with Prof. research and documentation of the his - Ronald de Wolf from the CWI Algorithms & Complexity research group, and Prof. Nikhil Bansal tory of an object as the contemporary from the department of mathematics and computer science of the Technical University digital interpretation in the sense of the Eindhoven. The position is funded through an NWO-TOP grant. history of the transformations of a dig - ital object. The conference invites par - Requirements ticipants to share their understanding of Candidates are required to have a completed PhD in mathematics and/or computer science, and an the shifting interpretations and uses of excellent research track-record. The ideal candidate will have a strong mathematical background the notion of ‘provenance’: and knowledge in several of the following topics: algorithms, complexity, algebra, combinatorial • In what ways can it, or can it not, optimization, semidefinite optimization, quantum information, communication and information facilitate grounding in ‘knowledge’? theory. The candidate should also have a taste in interdisciplinary research at the frontier between • What is the role of the museum as mathematics and computer science. Further needed qualifications for candidates include proven mediator of cultural heritage in facili - research talent and good academic writing and presentation skills. Candidates are expected to tating and establishing such prove - have an excellent command of English. nance and knowledge? Conference topics: Terms and conditions • Documentation & interdisciplinarity The terms of employment are in accordance with the Dutch Collective Labour Agreement for • Object documentation and analytical Research Centres (“CAO-onderzoeksinstellingen”). The gross monthly salary for an employee on resources a full time basis, depending on relevant work experience, ranges from € 3,409 to € 4,154. • Provenance of materials and tech - Employees are also entitled to a holiday allowance of 8% of the gross annual salary and a year- niques end bonus of 8.33%. CWI offers attractive working conditions, including flexible scheduling. The • Field research and object documenta - appointment will be for a period of one year, starting as soon as possible before September 2018. tion • Object documentation and archival Application resources Applications can be sent to [email protected]. We will take applications until getting our position • Oral tradition and witnessing infor - filled. All applications should include a detailed CV, a motivation letter describing research inter - mation & connection with objects ests and reasons for applying to this project, a research statement, and a short description of the • Methods of knowledge verification PhD dissertation or of the most relevant publications. The applicant should provide names of up and documentation of knowledge to three scientists who are acquainted with their previous academic performance and can write revision (“subjective” , “objective” reference letters. and other forms of evidence) • Documentation for target groups (e.g. More information special needs, etc) • Position announcement: https://kwz.me/hBF • Object information as historical • About the Networks and Optimization at CWI: source / evidence (ethics of prove - https://www.cwi.nl/research/groups/networks-and-optimization nance of information) • For CWI, please visit https://www.cwi.nl or watch our video “A Fundamental Difference” about working at CWI (https://www.cwi.nl/general/a-fundamental-difference). More information: • About the vacancy, please contact Prof. Monique Laurent, [email protected]. www.cidoc2018.com

ERCIM NEWS 112 January 2018 49 “Through a long history of successful research collaborations in projects and working groups and a highly-selective mobility programme, ERCIM has managed to become the premier net - ERCIM work of ICT research institutions in Europe. ERCIM has a consis - tent presence in EU funded research programmes conducting Membership and promoting high-end research with European and global impact. It has a strong position in advising at the research pol - After having successfully grown to icy level and contributes significantly to the shaping of EC become one of the most recognized ICT Societies in Europe, ERCIM has opened framework programmes. ERCIM provides a unique pool of membership to multiple member institutes research resources within Europe fostering both the career per country. By joining ERCIM, your development of young researchers and the synergies among research institution or university can established groups. Membership is a privilege. ” directly participate in ERCIM’s activities Dimitris Plexousakis, ICS-FORTH, ERCIM AISBL Board and contribute to the ERCIM members’ common objectives playing a leading role in Information and Communication Technology in Europe: benefits of Membership • Building a Europe-wide, open network of centres of excellence in ICT and As members of ERCIM AISBL, institutions benefit from: Applied Mathematics; • International recognition as a leading centre for ICT R&D, as member of the • Excelling in research and acting as a ERCIM European-wide network of centres of excellence; bridge for ICT applications; • More influence on European and national government R&D strategy in ICT. • Being internationally recognised both as ERCIM members team up to speak with a common voice and produce strategic a major representative organisation in its reports to shape the European research agenda; field and as a portal giving access to all • Privileged access to standardisation bodies, such as the W3C which is hosted by relevant ICT research groups in Europe; ERCIM, and to other bodies with which ERCIM has also established strategic • Liaising with other international organi - cooperation. These include ETSI, the European Mathematical Society and Infor - sations in its field; matics Europe; • Promoting cooperation in research, • Invitations to join projects of strategic importance; technology transfer, innovation and • Establishing personal contacts with executives of leading European research insti - training. tutes during the bi-annual ERCIM meetings; • Invitations to join committees and boards developing ICT strategy nationally and internationally; About ERCIM • Excellent networking possibilities with more than 10,000 research colleagues across Europe. ERCIM’s mobility activities, such as the fellowship programme, ERCIM – the European Research leverage scientific cooperation and excellence; Consortium for Informatics and • Professional development of staff including international recognition; Mathematics – aims to foster collaborative • Publicity through the ERCIM website and ERCIM News, the widely read quarter - work within the European research com - ly magazine. munity and to increase cooperation with European industry. Founded in 1989, ERCIM currently includes 15 leading research establishments from 14 European How to become a Member countries. ERCIM is able to undertake con - sultancy, development and educational • Prospective members must be outstanding research institutions (including univer - projects on any subject related to its field of sities) within their country; activity. • Applicants should address a request to the ERCIM Office. The application should inlcude: ERCIM members are centres of excellence • Name and address of the institution; across Europe. ERCIM is internationally • Short description of the institution’s activities; recognized as a major representative • Staff (full time equivalent) relevant to ERCIM’s fields of activity; organization in its field. ERCIM provides • Number of European projects in which the institution is currently involved; access to all major Information • Name of the representative and a deputy. Communication Technology research • Membership applications will be reviewed by an internal board and may include groups in Europe and has established an an on-site visit; extensive program in the fields of science, • The decision on admission of new members is made by the General Assembly of strategy, human capital and outreach. the Association, in accordance with the procedure defined in the Bylaws ERCIM publishes ERCIM News, a quar - (http://kwz.me/U7), and notified in writing by the Secretary to the applicant; terly high quality magazine and delivers • Admission becomes effective upon payment of the appropriate membership fee in annually the Cor Baayen Award to out - each year of membership; standing young researchers in computer • Membership is renewable as long as the criteria for excellence in research and an science or applied mathematics. ERCIM active participation in the ERCIM community, cooperating for excellence, are met. also hosts the European branch of the World Wide Web Consortium (W3C). Please contact the ERCIM Office: [email protected] In brief

CWI merges with NWo Institutes organisation Google Awards Grant

From 1 January 2018, the ERCIM member CWI has been merged with the NWO for fake News Institutes Organisation, NWO-I. The other research institutes in the Netherlands that joined the recent merger are ASTRON, NIOZ, NSCR and SRON. The Dutch insti - detection to foRtH tutes AMOLF, ARCNL, DIFFER and Nikhef were already part of NWO-I. and University of In 2015, the Netherlands Organisation for Scientific Research (NWO) started a tran - sition to a new organizational structure. The new NWO is intended to be more flex - Cyprus ible, more effective, and more focused on collaboration, which means it will be in a better position to respond to developments in science and society. In the new NWO As part of its Digital News Initiative structure there is a clear distinction between a granting organization, NWO, and a (DNI), Google announced a €150 million separate institutes organization: the Netherlands Foundation of Scientific Research innovation fund that supports innovation Institutes, or NWO-I for short. in Digital News Journalism. In its most recent round of funding, Google sup - After the merger, CWI will continue to fulfil its mission to conduct pioneering ported “Check-it: Visualizing fake news research in mathematics and computer science, generating new knowledge in these on social media”, a collaborative project fields and conveying it to industry and to society at large. .The general director of between FORTH and University of CWI, Jos Baeten, continues to be responsible for the CWI mission, for the day to day Cyprus. management of the institute, for the appointment of all CWI personnel and for the implementation of (research) policies. The foundation board of NWO-I will advise Check-it empowers users with the tools and monitor Baeten. they need in order to check whether the stories they read on-line are fake or not. For more information about the merger, visit the NWO-I website: In this way (i) users will be able to see if https://www.nwo-i.nl/en/ what they read is fake, and (ii) they may be reluctant to forward news which are known to be fake.

CWI hosts EIt digital’s New Innovation Space The main problem that this project tries to address is that when users view news CWI houses a new innovation space of partner EIT Digital, which was opened on 2 on social media they usually have little, November. With this new location in the financial heart of the Netherlands, EIT if any, means to decide whether the Digital wants to boost the development of FinTech, together with its partners in the information they see is real or fake. Netherlands and Europe. Although it is true that some social media allow users to report any fake Jos Baeten (CWI news they see, most of the social media Director) and out there do not provide such function - Patrick Essers ality. Therefore, they place all the (Node Director burden of deciding whether a story is EIT Digital in the fake or real on the end user. In this Netherlands) project we propose to change the way open EIT people consume digital news by Digital’s new empowering users with an automated Innovation Space ability to “check” whether a story is at CWI. fake or not. Towards this end, the project extends the web browser capa - bilities with a “check it” button (a browser plug-in) that will enable users to check whether a story is true or fake.

The new satellite location offers EIT Digital a strong base to help FinTech companies For more information, please contact and to develop new digital FinTech initiatives. The reason for opening a second EIT Evangelos Markatos Digital location in the Netherlands is a strongly felt wish of the EIT Digital’s partners ([email protected]), Bright Cape, CWI, ING Bank, and TNO who cooperate in various FinTech innova - Marios Dikaiakos tion activities within the pan-European ecosystem of EIT Digital. The Municipality ([email protected]) and of Amsterdam supports this and CWI is offering the new location. G Pallis ([email protected])

EIT Digital is a leading European innovation and education organization. Its mission is to foster digital technology innovation and entrepreneurial talent for economic growth and quality of life in Europe. It brings together entrepreneurs from a partner - ship of over 130 European corporations, SMEs, start-ups, universities and research institutes.

ERCIM NEWS 112 January 2018 51 Special theme: Quantum Computing

ERCIM – the European Research Consortium for Informatics and Mathematics is an organisa - tion dedicated to the advancement of European research and development in information tech - nology and applied mathematics. Its member institutions aim to foster collaborative work with - in the European research community and to increase co-operation with European industry.

ERCIM is the European Host of the World Wide Web Consortium.

Consiglio Nazionale delle Ricerche Norwegian University of Science and Technology Area della Ricerca CNR di Pisa Faculty of Information Technology, Mathematics and Electri - Via G. Moruzzi 1, 56124 Pisa, Italy cal Engineering, N 7491 Trondheim, Norway http://www.iit.cnr.it/ http://www.ntnu.no/

Centrum Wiskunde & Informatica Science Park 123, NL-1098 XG Amsterdam, The Netherlands RISE SICS http://www.cwi.nl/ Box 1263, SE-164 29 Kista, Sweden http://www.sics.se/

Fonds National de la Recherche 6, rue Antoine de Saint-Exupéry, B.P. 1777 L-1017 Luxembourg-Kirchberg http://www.fnr.lu/ SBA Research gGmbH Favoritenstraße 16, 1040 Wien http://www.sba-research.org/

Foundation for Research and Technology – Hellas Institute of Computer Science P.O. Box 1385, GR-71110 Heraklion, Crete, Greece FORTH http://www.ics.forth.gr/

Magyar Tudományos Akadémia Számítástechnikai és Automatizálási Kutató Intézet P.O. Box 63, H-1518 Budapest, Hungary http://www.sztaki.hu/

Fraunhofer ICT Group Anna-Louisa-Karsch-Str. 2 10178 Berlin, Germany http://www.iuk.fraunhofer.de/

University of Cyprus P.O. Box 20537 1678 Nicosia, Cyprus http://www.cs.ucy.ac.cy/ INESC c/o INESC Porto, Campus da FEUP, Rua Dr. Roberto Frias, nº 378, 4200-465 Porto, Portugal

Universty of Warsaw Faculty of Mathematics, Informatics and Mechanics Institut National de Recherche en Informatique Banacha 2, 02-097 Warsaw, Poland et en Automatique http://www.mimuw.edu.pl/ B.P. 105, F-78153 Le Chesnay, France http://www.inria.fr/

I.S.I. – Industrial Systems Institute VTT Technical Research Centre of Finland Ltd Patras Science Park building PO Box 1000 Platani, Patras, Greece, GR-26504 FIN-02044 VTT, Finland http://www.isi.gr/ http://www.vttresearch.com

Subscribe to ERCIM News and order back copies at http://ercim-news.ercim.eu/