NEST: the Neural Simulation Tool PARTNERS

Image source: Yury V. Zaytsev V. Yury source: Image

NEST: The Neural Simulation Tool PARTNERS

A SIMULATOR TO EXPLORE AND UNDERSTAND PROJECTS BIOLOGICAL NEURONAL NETWORKS.

Science has driven the development of the NEST simulator for the past 20 years. Originally created to simulate the propagation of synfire chains using single-processor workstations, we have pushed COMMUNITY BUILDING NEST‘s capabilities continuously to address new scientific questions and computer architectures. Prominent examples include studies on spike-timing dependent plasticity in large simulations of cortical networks, the verification of mean-field models, models of Alzheimer‘s and Parkinson‘s disease and tinnitus. Recent developments include a significant reduction in memory requirements, as demonstrated by a record-breaking simulation of 1.73 billion neurons connected by 10.4 trillion synapses on the Japanese K supercomputer, paving the way for brain-scale simulations. QUALITY AND PERFORMANCE Running on everything from laptops to the world‘s largest supercomputers, NEST is configured and controlled by high-level Python scripts, while harnessing the power of C++ under the hood. An extensive testsuite and systematic quality assurance ensure the reliability of NEST.

APPLICATIONS

USAGE EXAMPLE PARTNERS RUHR-UNIVERSITÄT BOCHUM (RUB) PARTNERS NEST originated as a student project at the Institute for Neural Compu- tation over 20 years ago. Today, the group Functional Neural Circuits at RUB is investigating top-down models of network function and adapting NEST accordingly.

OKINAWA INSTITUTE OF SCIENCE AND TECHNOLOGY (OIST) NORWEGIAN UNIVERSITY OF LIFE SCIENCES (NMBU) The development of NEST OIST is developing large-scale models of the cortex, the thalamus, and the The Computational Neuroscience Group at the Department of Mathematical Sciences and Techno- is driven by the demands basal ganglia using NEST in the context of a natio- logy (IMT) at NMBU has made key contributions to NEST‘s hybrid parallelization and nal strategic program on the K supercomputer. of neuroscience and is lead developer for the Topology Module for spatially structured networks. Com- carried out in a bining NEST with the simulator NEURON and the toolbox LFPy, it develops multi- ADVANCED INSTITUTE FOR COMPUTATIONAL SCIENCE (AICS) collaborative fashion scale models predicting LFP and MUA signals from network simulations. at different institutes all The AICS in Kobe is collaborating with Forschungs- over the world. NATIONAL BERNSTEIN NETWORK COMPUTATIONAL NEUROSCIENCE zentrum Jülich on the development of NEST for the K supercomputer and future exascale systems. Researchers from the Bernstein Center Freiburg (BCF) and the “Bernstein Facility New functionality is always added by Simulation and Database Technology” (BFSD) at Forschungszentrum Jülich work researchers who need it, ensuring the on neuron models and infrastructure for developing and deploying NEST. FORSCHUNGSZENTRUM JÜLICH GMBH / RWTH AACHEN UNIVERSITY / JÜLICH AACHEN RESEARCH ALLIANCE most usable implementation of novel Virtual Reality Group of RWTH and features. Neuroscientists at the Institute of Neuroscience and Under the umbrella of the Jülich Aachen Research INM develops methods for inter- Medicine (INM) at Forschungszentrum Jülich use NEST Alliance (JARA), several cooperations between For- active visualization of brain-scale ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE (EPFL) To guarantee the best possible per- for large-scale brain simulations on the world‘s biggest schungszentrum Jülich and RWTH Aachen Univer- simulation data. The Simulation formance of the code and a maximum The Neurorobotics Section of the EPFL Blue Brain Project uses and develops NEST for neuro- supercomputers. Together with neuroinformatics specia- sity significantly contribute to the development of Laboratory Neuroscience (SLNS) accuracy of the scientific results ob- robotics simulations with closed action-perception loops. In addition, NEST is used to develop and lists at the INM and experts at the Jülich Supercomputing NEST. The collaboration with the High Performance provides infrastructure for quality tained, neuroscientists are supported investigate data-driven whole-brain models of rodents at the level of point Centre (JSC) they develop high-performance algorithms Computing (HPC) Group at RWTH optimizes NEST assurance and continuous integra- by experts in computational scien- neurons. Future work will focus on extending the capabilities of NEST towards and memory efficient data struc- for HPC systems. A col- tion for NEST. ces, software engineering, and math. the real-time control of simulated and physical robots. tures for NEST. laboration between the

TI EC ON

TI NN

CO F CORTICAL-LIKE O PROJECTS ATION NET WOR UR KS MU IG ONF E PERCEPTION

SI V ION, C CTI E E R ING A

D R CT AL U SC D TI- BRAINSCALES

UL EN-LOOP DYNAMICS G AN FUN M S & OP TATION

E, IN ESEN

REPR LS OF SENSORY CORTICES

LL E UR G OD GIN ALE M MER TI-SC E MUL RGE SIMULATION AND MODELING MODE A

N L STRUCT I N S

THOD N N OLOGY E AI N M N TECH NT OF FOR THE HUMAN BRAIN

AI ATIO LOPME HUMAN BRAIN PROJECT BRAINSCALES

BR MUL DEVE

BR SI

T Y IN Since 2013 the Helmholtz Association The EU Flagship Human Brain Project (HBP) star- BrainScaleS is a four-year multi-center and multi- I T

IL Y

FAC U G

TISCALE L O

HYBRID MU MP

AR E L O has funded the portfolio theme “Super- ted in October 2013. disciplinary EU-funded project investigating the PROJECTS TW C

SOF O I

TING D

R E

O B

SUPP S

A computing and Modeling for the Human

B brain’s multi-scale nature to advance neuro-

N R IO -

AT U T KE U I

MP E

CO P N

AL S Brain” (SMHB), Today, there is an explosion of data on the brain R

GENERIC NEU morphic hardware and brain-inspired computing. E

A AL

T - which at the end with 100,000 articles published in 2012 alone. O

N E

E I

M G

OP T

EL G

V A

RE DE A H A T ARDW N L

I of the initial 5 However, there is no method available to inte-

U Research covers fields from electrophysiology

O O

N I

M C years funding grate these data. The HBP plans to build up the

C and computational neuroscience to chip design

N CT

O D

T C

I I

O N

T D

L will merge into technical infrastructure for modeling and simula-

A I and theoretical compu- W

SIMULATION AND MODELING S

L H

R NE

M P

FOR THE HUMAN BRAIN I E O

E ScaleS

T S

C the then fully ting the whole brain. In strategy and scope this is

I ter science. BrainScaleS

O S

U Y N

M IE

I S

E C

N ScaleS

S Brain I

N operational new Jülich research program like building the Very Large Telescope (VLT) of the

R R N

A and its precursor FACETS I

U A B R

E R B B

N F “Decoding the Human Brain” and the Euro- ESO in Chile which enables us to reach out into have introduced large-

M E S O

ATH UR IC ScaleS T

EM OI A M

ATIC N M L

AL FO OR A

AND R F peanE Human Brain Project. previously unexplored territory. This requires the scale collaborative sci-

T M N R HE A I

ORE T AL

TIC IC C HE

A I T

L FO S D

UND E D joint effort of many laboratories across Europe ence into neuroscience, now feeding the Human AT M N

IONS

OF YO

BRA IONS E I T B N A RES PLIC NEST is the main simulation platform used and hundreds of researchers. Brain Project, and established Europe‘s leader- EA RCH AP The complexity of modern neuroscience and developed in the SMHB. ship in neuromorphic computing. For neuroscience where researcher are used to ES RCHIT ECTUR COGNIT IVE A increasingly requires the formation A strategic measure of the SMHB is the working in small groups the project is expected In BrainScaleS, NEST is used for research on HUMAN ATA of large and long-term research consortia foundation of the Simulation Laboratory to induce a cultural change. NEST is the simu- full-scale cortical network models as well as on N D BRAIN PROJECT RAI N B Neuroscience integrated into the national lation engine developed and used in the HBP at MA and a collaborative work style. abstract models of network dynamics and ser- U TA C H EGI DA Bernstein Network Computational Neuro- the level of resolution of neurons and synapses. AT IN ves as the reference platform for the hardware TR A S R B E The NEST Initiative develops strategic plans with those consortia to enable collaboration on science as the “Bernstein Facility for systems HMF and SpiNNaker under develop- US MO network models, ensures the availability of reliable and reproducible simulation technology, Simulation and Database Technology” in IC ment in Heidelberg and Manchester, respec- G TE RA and safeguards the sustainability of associated investments. Germany. tively. ST COMMUNITY BUILDING

PUBLICATION OF METHODS AND TECHNOLOGIES LICENSING USER SUPPORT TEACHING AND TUTORIALS

The NEST Initiative promotes the development of We have provided public releases of NEST‘s One important aspect of the success of NEST is NEST is one of the core simulators taught at a culture of collaborative neuroscience through source code to the computational neuroscience its user community. 300+ users are subscribed the main computational neuroscience summer scientific talks, publications, and the activities of community free of charge since 2004 under the to the user mailing list, where questions about schools, i. e. the Advanced Course for Computa- its members. The transfer of computer science proprietary NEST License. using and extending NEST are discussed along- tional Neuros- knowledge to neuroscience is a crucial aim. side general modeling questions. cience (ACCN), Since 2012, NEST is available under the GNU the Okinawa In recent years, members of the NEST core develop- General Public License version 2 or later as a NEST developers are subscribed to this list to Institute of

ment team have published many articles about no- consequence of popular demand by commu- provide high- Technology‘s COMMUNITY BUILDING vel methods and technologies for neural simulation. nity members. The main benefit of this change quality answers Computatio- Prominent examples are the hybrid parallel and dis- is that the new license ensures the free availa- and support nal Neurosci- The NEST Initative was founded in 2001 to foster scientific tributed execution of simulations, the user interface bility of NEST and allows users with pro- ence Course layer for NEST, the improvement of the performance the users to re-distribute blems that re- (OCNC), and the Latin American School on Com- exchange of computational neuroscience methods and to and reduction of the modified versions of NEST quire a deeper putational Neuroscience (LASCON). provide a framework for joint NEST development, support and memory usage, and to fellow researchers. insight into the NEST architecture. In recent documentation to benefit the scientific community. suggestions for the years, however, more and more questions get In addition, we give tutorials on advanced topics description of neuro- In addition to NEST answered by users themselves, which shows around NEST at the annual Computational Neuro- The NEST Initiative promotes NEST, e.g., The NEST Initiative is incorporated as a non- nal network models in licensing, we support the liveliness of our community. science Meeting (CNS) to provide users with through the NEST website and mailing lists, profit organization in Ecublens, Switzerland, publications. licensing discussions for updates on current developments and direct and distributes NEST code under open source and is managed by a board of directors. Any- other scientific tools to In addition to this interactive help, we provide contact with NEST developers. Recent tutorials licenses. one who wants to support our work on NEST Currently, over 290 assure good availability and allow the inspection a documentation index of the NEST simulator, featured topics such as the general modeling may join the Initiative as a community mem- papers cite NEST as of the tools by others. These are important cor- which explains general modeling questions, approach in NEST, creating structured large- NEST‘s development is coordinated through a ber. Active developers may become active the simulator used in nerstones to reproducible science which foster and a list of examples that help users to get scale network models, and developing neuron monthly developer video conference. members with full voting rights. modeling studies. the reuse of code. started quickly with NEST. and synapses models for NEST. QUALITY AND PERFORMANCE

TESTSUITE CONTINUOUS INTEGRATION NEST is a high performance neuronal network simulator, developed with a focus on reproducibility and correctness. We employ an extensive testsuite to ensure the correct compila- To guarantee the quality of NEST development, we use continuous integration (CI), tion and installation of NEST on the user‘s computer. i.e. a practice where quality assurance (QA) is applied continuously as opposed to the traditional procedure, where QA is only applied during the integration phase after PERFORMANCE AND MEMORY Testing is a standard activity in computer science to make sure that completing all development. CI decreases the risks associated with the integration all components of an application work as expected and that bogus by spreading required efforts over time, which helps to improve software quality and To ensure the best possible use of available computing resources, we are cons- Performance and memory consumption user input leads to the correct error condition. to reduce the time taken to deliver it. tantly improving the performance and memory footprint of NEST. A considerable part of the recent software development has been directed towards the re-design The testsuite of NEST consists of a set of hierarchical small unit A Jenkins-based CI infrastructure (Zaytsev and Morrison, 2012, doi:10.3389/ of NEST in order to meet the memory challenges of brain-scale simulations on ZE tests, organized from simple to complex: fninf.2012.00031) helps with regular and automated testing of new commits and re- supercomputers, while keeping it slick and simple for laptop use (Helias et al. QUALITY AND PERFORMANCE RK SI ports identified prob- 1. Test NEST‘s ability to report errors Continuous integration workflow 2012, doi:10.3389/fninf.2012.00026). WO lems such as failed DEVELOPER 1

NET 2. Test that neurons and synapses have correct defaults and broken tests in a The resulting 3rd generation simulation kernel (3g) as released with NEST 2.2 3. Test that neurons and synapses accept new parameters timely fashion. CONTINUOUS INTEGRATION and the novel 4g kernel scale well in terms of memory usage and runtime on con- 4. Compare simulation results with analytical results DEVELOPER 2 VCS temporary supercomputers such as the JUQUEEN BlueGene/Q machine (JQ, blue 5. Test the parallelization features of NEST This technology also markers) and the K computer (red markers). ME (S) allows us to create re-

M TI 6. Test the high-level user interface functions

SI gular public releases WWW Profiling and modeling of NEST’s resource consumption and run-time behavior are In addition, we use regression tests to ensure that bugs that were of NEST from the inter- RAPID FEEDBACK essential tools for this endeavour. NUMBER OF VIRTUAL PROCESSES fixed don‘t come back in newer versions. nal revision.

Alzheimer‘s disease. Alzheimer‘s architectures. library), NEST can enrich the repertoire of hardware simulations. hardware of repertoire the enrich can NEST library),

ral networks subject to a disruption of connectivity, a hallmark of of hallmark a connectivity, of disruption a to subject networks ral to serve as a natural source of uncorrelated “noise” for functional neural neural functional for “noise” uncorrelated of source natural a as serve to By pre-processing data or through online interactions (e.g. using the MUSIC MUSIC the using (e.g. interactions online through or data pre-processing By

- neu of performance classification and sensitivity the restore fully ministic neural networks networks neural ministic gate model dynamics at different scales under hardware-specific conditions. conditions. hardware-specific under scales different at dynamics model gate

2202-14-S1-P282) have shown that homeostatic mechanisms can can mechanisms homeostatic that shown have 2202-14-S1-P282) hs fet los deter- allows effect This Harware-specific constraints can be implemented in NEST in order to investi to order in NEST in implemented be can constraints Harware-specific -

APPLICATIONS

s on ur ne ic minist ter de

ih h hl o NS, ahan t l (03 doi:10.1186/1471- (2013, al. et Bachmann NEST, of help the With

of rk o etw n ed gn si de

work activity. work the realm of conventional computing architectures. computing conventional of realm the RK WO NET

NAL NAL O CTI FUN

- net neural decorrelate difficulties arise both on the technical and conceptional level when leaving leaving when level conceptional and technical the on both arise difficulties

(mV) HT G WEI IC SYNAPT

networks, can efficiently efficiently can networks, conventional computers in terms of speed and energy efficiency. However, However, efficiency. energy and speed of terms in computers conventional

w lo

dant in biological neural neural biological in dant DI surpass can devices these architectures, parallelized highly their to Due

e is no

SCR EXC.-EXC. EXC.-EXC.

- abun feedback, bitory

neural networks. These substrates can either use digital or analog circuits. circuits. analog or digital use either can substrates These networks. neural

sis ta IMINAT homeos

Simulation visualization visualization Simulation

- inhi that shown recently Brain-inspired hardware devices provide a novel substrate for simulating simulating for substrate novel a provide devices hardware Brain-inspired I E

IO ment in cooperation with the Virtual Reality Group of RWTH Aachen University. Aachen RWTH of Group Reality Virtual the with cooperation in ment pcbi.1002596) have have pcbi.1002596)

IN PAN CA BI

REFERENCE FOR NEUROMORPHIC HARDWARE NEUROMORPHIC FOR REFERENCE

tion results, an interactive visualization tool called VisNEST is under develop under is VisNEST called tool visualization interactive an results, tion -

doi:10.1371/journal. -D

loss e naps sy

s on ur ne c ti inis rm te de of rk o etw n ndom ra EG

synapses. To display and analyze the simula the analyze and display To synapses. 2·10 via connected neurons Tetzlaff et al. (2012, (2012, al. et Tetzlaff - RK WO NET SE

NOI REE

and dysfunction. and macaque visual cortex and has been successfully run with up to 20 million million 20 to up with run successfully been has and cortex visual macaque and mathematical tools, tools, mathematical and TY

gh hi

To study large-scale interactions, the model was extended to all areas of the the of areas all to extended was model the interactions, large-scale study To With the help of NEST NEST of help the With neuroscience and to investigate cortical function function cortical investigate to and neuroscience Network sketch Network

Classification performance Classification NEST is used for diverse applications in computational computational in applications diverse for used is NEST

This model accounted for various aspects of spontaneous cortical activity. activity. cortical spontaneous of aspects various for accounted model This ground noise. So far, the biological origin of this noise remains unclear. remains noise this of origin biological the far, So noise. ground

changes at the cellular, dynamical and behavioural level. behavioural and dynamical cellular, the at changes sensory cortex (Potjans and Diesmann, 2012, doi:10.1093/cercor/bhs358). doi:10.1093/cercor/bhs358). 2012, Diesmann, and (Potjans cortex sensory network models, for example, rely on the presence of uncorrelated back uncorrelated of presence the on rely example, for models, network -

NEST has been used to develop a full-scale model of a 1mm a of model full-scale a develop to used been has NEST disease-related between relationship the about knowledge our ses lenge to computational neuroscientists. A number of powerful functional functional powerful of number A neuroscientists. computational to lenge early of patch

- increa This performance. computational and dynamics network on - chal major a constitutes often ingredients realistic biologically using

structured networks ranging from cellular circuits to sets of interacting areas. interacting of sets to circuits cellular from ranging networks structured diseases neurodegenerative of effects the investigate to us enable realization in the neural hardware. Implementing functional algorithms algorithms functional Implementing hardware. neural the in realization

Cognitive and sensual processes in the mammalian cortex are supported by by supported are cortex mammalian the in processes sensual and Cognitive function brain implementing models network realistic Biologically A central goal of neuroscience is to understand brain function and its its and function brain understand to is neuroscience of goal central A

CORTICAL MODELING CORTICAL DISEASES NEURODEGENERATIVE BRAIN-INSPIRED COMPUTING BRAIN-INSPIRED APPLICATIONS

USAGE EXAMPLE EXAMPLE USAGE

along the horizontal axis, neuron index increases along the vertical axis. axis. vertical the along increases index neuron axis, horizontal the along MPI-PARALLELISM to use computer clusters and supercomputers supercomputers and clusters computer use to MPI-PARALLELISM ·

S) (M ME TI

shows the spiking activity of 50 neurons as a raster plot. Time increases increases Time plot. raster a as neurons 50 of activity spiking the shows

MULTI-THREADING to use multi-processor machines efficiently machines multi-processor use to MULTI-THREADING ·

0 10 90 80 70 60 50 40 30 20 10

dom network with recurrent excitation and inhibition. The figure to the right right the to figure The inhibition. and excitation recurrent with network dom Parallelization: RA

The example script on the left simulates spiking activity in a sparse ran sparse a in activity spiking simulates left the on script example The -

Backend for the simulator-independent modeling tool PyNN PyNN tool modeling simulator-independent the for Backend

· Z)

0 Interface to the Multi Simulator Coordinator MUSIC Coordinator Simulator Multi the to Interface ·

and synapse models. synapse and

TOPOLOGY MODULE for creating complex spatially structured networks structured spatially complex creating for MODULE TOPOLOGY ·

with programming experience can also extend NEST‘s repertoire of neuron neuron of repertoire NEST‘s extend also can experience programming with

Built-in SIMULATION LANGUAGE INTERPRETER (SLI) (SLI) INTERPRETER LANGUAGE SIMULATION Built-in ·

be combined with Python‘s many packages for scientific computing. Users Users computing. scientific for packages many Python‘s with combined be

20 PYTHON BASED USER INTERFACE (PyNEST) INTERFACE USER BASED PYTHON · NE

UR at many levels. As a module to the interpreted language Python, NEST can can NEST Python, language interpreted the to module a As levels. many at

Interfaces: ON

30 ID extensible is NEST supercomputers. largest world‘s the to computers laptop

EXTENSIVE TESTSUITE for continuous verification of the code base code the of verification continuous for TESTSUITE EXTENSIVE ·

NEST is designed to optimally use the available computing hardware, from from hardware, computing available the use optimally to designed is NEST

40 GRID BASED NEURONAL INTERACTION and INTERACTION IN CONTINUOUS TIME CONTINUOUS IN INTERACTION and INTERACTION NEURONAL BASED GRID ·

EXACT INTEGRATION is used for suitable neuron models neuron suitable for used is INTEGRATION EXACT ·

devices that help stimulating the network and observing its response. its observing and network the stimulating help that devices 50

Each neuron model uses an APPROPRIATE SOLVER APPROPRIATE an uses model neuron Each ·

a large set of tested models for neurons and synapses as well as virtual virtual as well as synapses and neurons for models tested of set large a

Simulation results Simulation

Accuracy and verification: and Accuracy

synaptic plasticity as observed for example during learning. NEST provides provides NEST learning. during example for observed as plasticity synaptic

the mean firing-rate of the neurons. the of firing-rate mean the from neurons and their connections, which can either be static, or exhibit exhibit or static, be either can which connections, their and neurons from STATIC AND PLASTIC SYNAPSE MODELS (STDP, STP, dopamine) STP, (STDP, MODELS SYNAPSE PLASTIC AND STATIC ·

given time. The lower part of the figure shows a histogram with with histogram a shows figure the of part lower The time. given riment, but takes place inside a computer. The neural network is generated generated is network neural The computer. a inside place takes but riment, Simple MULTI-COMPARTMENT NEURON MODELS NEURON MULTI-COMPARTMENT Simple ·

Each dot corresponds to a spike of the respective neuron at a a at neuron respective the of spike a to corresponds dot Each A NEST simulation follows the logic of a laboratory electrophysiological expe electrophysiological laboratory a of logic the follows simulation NEST A - HODGKIN-HUXLEY TYPE NEURON MODELS with one compartment one with MODELS NEURON TYPE HODGKIN-HUXLEY ·

MAT2 NEURON MODEL (Kobayashi et al., 2009) al., et (Kobayashi MODEL NEURON MAT2 ·

or models of network activity dynamics. activity network of models or

ADAPTIVE-EXPONENTIAL IAF (Brette and Gerstner, 2005) Gerstner, and (Brette IAF ADAPTIVE-EXPONENTIAL ·

processing, models of learning and plasticity, plasticity, and learning of models processing, based synapses based

conductance- and current- with MODELS NEURON (IAF) FIRE AND INTEGRATE · such as models of visual or auditory information information auditory or visual of models as such

Neuron and synapse models: synapse and Neuron

NEST is a simulator for spiking neural network models models network neural spiking for simulator a is NEST

Simulation script Simulation Key features of of features Key NEST USAGE EXAMPLE USAGE ACKNOWLEDGEMENTS

Over the years, NEST‘s development has been supported by the Weizmann Institute, University Bochum, Universi- ty & BCCN Freiburg, Honda Research Institute Europe, MPI for Fluid Dynamics, Norwegian University of Life Scien- ces, RIKEN Brain Science Institute, Helmholtz Association and Forschungszentrum Jülich, EPFL and BlueBrainPro- ject, EU grants FACETS (FP6-15879), BrainScales (FP7- 269921) and Human Brain Project (604102), Research Council of Norway grant eNeuro (178892/V30) and JARA. Special thanks for contributing to this brochure go to The development of NEST is coordinated by the Sacha van Albada, Michaela Bleuel, Paul Chorley, David NEST Initiative, a non-for-profit organization Dahmen, Markus Diesmann, Jochen Martin Eppler, Marc- open for interested scientists. NEST is availa- Oliver Gewaltig, Bernd Hentschel, Jakob J. Jordan, Susan- ble as free software under the terms of the ne Kunkel, Abigail Morrison, Christian Nowke, Alexander GNU General Public Peyser, Hans Ekkehard Plesser, Martina Reske, Maximili- License to ensure wide dissemination and allow researchers an Schmidt, Jannis Schuecker, Tom Tetzlaff, and Yury V. to adapt NEST to their needs. Zaytsev. More information about NEST as well as its source code is In particular, we gratefully acknowledge the Jülich available on the homepage of the NEST Initiative. Aachen Research Alliance (JARA) for the realization of http://www.nest-initiative.org this brochure.