Sep/Oct 2021
Total Page:16
File Type:pdf, Size:1020Kb
CERNSeptember/October 2021 cerncourier.com COURIERReporting on international high-energy physics WELCOME CERN Courier – digital edition Welcome to the digital edition of the September/October 2021 issue of CERN Courier. ARTIFICIAL As data volumes surge, deep learning is becoming increasingly important in INTELLIGENCE particle physics. This special edition on artificial intelligence (AI) captures two new trends: using “unsupervised” deep learning to spot anomalous events, and designing AI that can “think not link”. Community-organised data challenges are leading the way (p27) and deep learning could even be used in the level-one triggers of LHC experiments (p31). To keep up with the cutting edge of AI research, physicists are reaching out to computer science and industry (p36): the latest developments could help explore theory space (p51) and build trust in AI to do more of the heavy lifting throughout the analysis chain (p49). We also explore recent thinking that an ordered simplicity may emerge from the complexity of deep learning in a similar way to statistical mechanics and quantum field theory (p39). Elsewhere in the issue: a tribute to Steven Weinberg (p65); a SciFi upgrade for LHCb (p43); reports from the summer conferences (p19); the most stable tetraquark yet (p7); quantum gravity in the Vatican (p59); anisotropies point to cosmic-ray origins (p11); and much more. To sign up to the new-issue alert, please visit: http://comms.iop.org/k/iop/cerncourier To subscribe to the magazine, please visit: https://cerncourier.com/p/about-cern-courier Emergence Deep learning in the trigger EDITOR: MATTHEW CHALMERS, CERN DIGITAL EDITION CREATED BY IOP PUBLISHING The most stable tetraquark yet CCSepOct21_Cover_final_v1.indd 1 04/08/2021 11:50 CERNCOURIER www. V OLUME 6 1 N UMBER 5 S EPTEMBER / O CTOBER 2 0 2 1 Your DIGITAL CERNCOURIER.COM POWER ELECTRONICS Partner IN THIS ISSUE V OLUME 61 N UMBER 5 S EPTEMBER /O CTOBER 2 02 1 It’s ALL Rayner Sparks!/M iStock/georgeclerk LHCb about the DETAILS AI leaders Experts weigh in on what’s next Order in complexity A surprising link between SciFi upgrade LHCb prepares for a in the world of artificial intelligence.36 theoretical physics and AI research. 39 luminous future. 43 NEWS PEOPLE ANALYSIS ENERGY FRONTIERS FIELD NOTES CAREERS OBITUARIES Near stable tetraquark Bs decays remain AI centre stage at Quantum gravity Steven Weinberg • SuperKEKB raises bar anomalous • Charm CHEP • FCC Week • LHCP in the Vatican • Felix H Boehm • Science Gateway • DUNE breaks fragmentation • Strange Quark Matter Residents of the Vatican • Anatoly Efremov cryostats • Surveying universality • Unbalanced • Long-lived particles Observatory describe life • Alexei Onuchin a future collider photon pairs • Machine • IPAC • Sustainable as a full-time physicist • Tord Riemann. 65 • Web code auctioned learning and top-quark HEP. 19 in the church. 59 • Cosmic-ray origins. 7 interactions. 15 FEATURES DATA CHALLENGES DEEP LEARNING SCIENCE AND INDUSTRY THEORETICAL PHYSICS LHCb UPGRADE What’s in the box? Hunting anomalies Forging the future Emergence Building the future Grassroots initiatives with an AI trigger of AI Erik Verlinde sizes up of LHCb like the LHC Olympics How ‘unsupervised’ The Sparks! forum will the Standard Model, The experiment boasts and Dark Machines learning could probe gather machine-learning gravity and intelligence a brand-new “SciFi” stimulate innovation every LHC event for experts to encourage as candidates for future tracker and upgraded in AI. 27 signs of new physics. 31 innovation in particle explanation as emergent RICH detectors. 43 physics and AI. 36 phenomena. 39 OPINION DEPARTMENTS VIEWPOINT INTERVIEW REVIEWS FROM THE EDITOR 5 Power Supply Systems Designing an Stealing theorists’ An enduring and NEWS DIGEST 13 Precision Current Measurements AI physicist lunch adaptable design LETTERS 53 Jesse Thaler argues that Anima Anandkumar and Resistive gaseous APPOINTMENTS 60 particle physicists must John Ellis explore how detectors Climate Beamline Electronic Instrumentation • CMS-OUTREACH-2019-001/M Rayner & AWARDS go beyond deep learning AI could collaborate with change and energy On the cover: How deep learning could help find new RECRUITMENT 61 FMC and MicroTCA and design AI capable of theoretical physicists. 51 options for a sustainable deep thinking. 49 future. 55 physics at the LHC. 31 BACKGROUND 70 www.caenels.com CERN COURIER SEPTEMBER/OCTOBER 2021 3 CCSepOct21_Contents_v3.indd 3 04/08/2021 12:12 MASTER - February 2021_Paolo.indd 1 03/08/21 12:20 CERNCOURIER www. V OLUME 6 1 N UMBER 5 S EPTEMBER / O CTOBER 2 0 2 1 CERNCOURIER.COM FROM THE EDITOR MMS New trends in AI for particle physics eep learning has transformed the world over the past S Sion i/M Pier i n i THE UNIVERSAL SYSTEM decade. As data volumes surge, it is also growing D in importance in high-energy physics (HEP). This FOR A LARGE RANGE OF LOW AND HIGH VOLTAGE special edition captures two new trends: using “unsupervised” learning to boost creativity in theory building, and designing ar t ificial intelligence (AI) t hat can “t hin k not lin k”. MULTICHANNEL POWER SUPPLIES Most deep learning is “supervised”. Layer upon layer of Mark Rayner artificial neurons operate on data, abstracting electrical signals Deputy editor into particles and jets, and then event topologies, until the data are compressed into a single label: the Standard Model, or whichever flavour of new physics you are searching for. So far, the stack of new-physics labels has never come close to d i s c o v e r y s i g n i fic a n c e, a n d y e t t h e d e fic i e n c i e s i n t h e St andrd Model are there for all to see. What if we’re looking for the wrong new physics? This situation calls for unsupervised learning, where NS? ASK you don’t need to tell the machine what to look for. Just 40 MHz A data challenge to design artificial intelligence for IO E one unsupervised analysis has been published so far, and the level-one trigger has begun. T R S I community-organised data challenges are leading the way E C H U (p27). But supervised triggering algorithms mean that LHC t h at t h i n k s l i ke a p hy sic i s t i f we a re to t r u s t it to d o more a nd ! Q a n a l y s e r s a r e a l r e a d y l o o k i n g a t a s t a c k e d d e c k o f d a t a a d a p t e d more of the heavy lifting in particle physics (p49). to the needs of the most popular theories. To unlock the full But with all its inscrutable complexity, is AI really a power of unsupervised learning, it must be unleashed on the suitable tool for fundamental science? A recent current of W-IE-NE-R + ISEG f u l l 40 MHz data flow in the level-one triggers (p31). Readers opinion (for example, J Halverson et al. 2021 Mach. Learn.: Sci. Technol. 2 035002) asserts that deep learning may actually SUPPORT OFFICE AT CERN w h o w a nt t o g e t i n v ol v e d c a n t a k e p a r t i n a n e w d a t a c h a l l e n g e launched at the ML4Jets conference in July – you have a year exhibit emergent simplicity in a similar way to statistical Dr. Dipl.-Phys. Erich Schaefer to submit your algorithms. m e c h a n i c s a n d q u a nt u m fie l d t h e or y, w h o s e l a w s a r e n o t fun- Bat.: 13/R-023 | +41 75411 4301 | [email protected] damental but emerge from complexity via a process of data We need to Think not link compression. To e x plore t h i s f a sc i n at i ng l i n k to A I resea rc h, MMS MMS design AI that At their core, deep-learning algorithms are association we asked Erik Verlinde to size up the Standard Model, grav- HV LV thinks like a machines, with limited intelligence. Today, AI research is ity and intelligence as candidates for future explanation as compatible compatible physicist if we dominated by the question of how to think rather than link. emergent phenomena (p39). are to trust it to To make sure HEP is part of the conversation, CERN has invited Elsewhere in this issue: a tribute to Steven Weinberg (p65); EBS HV BOARDS HIGH CLASS 50 world leaders on AI to the inaugural Sparks! Serendipity a S c i F i up g r a d e for L HC b (p43); a n i s o t r o pie s p oi nt t o c o s m ic-r a y do more of the CONTROLLER BOARDS WITH LINUX INSIDE Bipolar 4 quadrant high voltage LOW VOLTAGE BOARDS Forum in September (p36). AI initiatives are springing up or i g i n s ( p1 1); t h e m o s t s t a bl e t e t r a q u a r k y e t ( p 7); r e p or t s f r o m heavy lifting Integrated EPICS IOC, SNMP, OPC-UA module with CFG.