NEST SpiNNaker Elephant Demo

The HBP Collabarotary is constantly developed and updated. Meanwhile some tools used in this Collab are outdated, such as the task framework and some notebooks. Parts of this overview document is still referring to these deprecated tools. We did not remove the outdated documentation and tools, since there is no possibility to track the progress (e.g. via version control). Instead we marked the old part as Version 1 to keep the history. Version 2 is built on top of the results achieved with Version 1 and shows new progress.

Version 1

This NEST SpiNNaker Elephant Demo Collab demonstrates how to assemble a workflow including modeling, simulation, analysis and visualization inside in the HBP Collaboratory.

Different approaches to simulating spiking neuronal networks are being pursued in the HBP. On the one hand, traditional simulations using highly scalable simulation software running on high-performance computers, such as the NEST simulator. On the other hand, 2 neuromorphic hardware platforms are developed within HBP, one of these being the SpiNNaker hardware system. Given the diversity of different simulation engines, one may ask: To what extent are the results they produce indeed comparable? In this Collab, we demonstrate one possibility of how the HBP Collaboratory can assist in approaching this question by enabling us to...

1. simulate a specific model using the NEST simulator on the JUQUEEN super computer
2. simulate the exact same model using the SpiNNaker hardware
3. pool the data in a central location accessible to all collaborators
4. validate the results by methods provided by Elephant, and
5. visualize the results.

Figure 1 shows how these steps make up an integrative loop, an iteration of data generation, data analysis,  and visualization aimed at validating the results from the different simulation engines against one another. It is in these iterative loops that we conceive collaborative work that brings scientists from different fields together.

A large-scale neural network simulation of a cortical microcircuit ( Potjans, T. C., & Diesmann, M. (2014) Cerebral Cortex 24(3):785-806 , see below) is run on the supercomputer JUQUEEN using the NEST simulator and on the neuromorphic hardware SpiNNaker in order to validate and compare simulation output. The resulting data is collected in the Collab and analyzed with a battery of statistical methods using the Elephant library. A first insight into the data can be obtained in an interactive fashion in a Jupyter Notebook, but more complex analysis is carried out using the JURECA HPC system. UNICORE links the Collaboratory with HPC facilities and allows an easy access to distributed computing. Final results of the analysis can be visualized by the visualization framework ViSTA.

A detailed description of this Collab can be found in Getting Started.

Note that access to HPC sites requires to undergo an application procedure for computing time. For getting test access to JUQUEEN and JURECA please contact HBP-HPC-Platform@fz-juelich.de .

Figure 1: Integrative loop linking modeling, analysis and visualization embedded into the collaborative framework of the HBP.

Cortical Microcircuit Model

The microcircuit simulation, which represents the modeling component of this Demo Collab, is a full scale neuronal network model. Full scale means that the model represents a particular biological circuit with neurons and synapses at their natural density. The model represents 1 mm³ of cortex and contains around 100,000 spiking point-neurons connected by around 1 billion synapses in four cortical layers. Each layer contains an excitatory and an inhibitory neuron population which are interconnected with cell-type and layer specific connection probabilities derived from experimental data of early sensory cortex. A sketch of the model is depicted in Figure 2.

Figure 2: Illustration of the multi-layered microcircuit model ( adapted from Potjans, T. C., & Diesmann, M. (2014) ).

Version 2

This part focuses mainly on the NEST-SpiNNaker Validation Notebook and shows a specific use case of the NetworkUnit framework. This framework is built on top of the Elephant library and demonstrates sophisticated methods to validate/compares spiking activity of simulated data coming from the NEST and SpiNNaker simulators.

Collaborative work of

• Coordination - Boris Orth, Anna Lührs
• Scientific-Coordination - Sonja Grün, Markus Diesmann, David Lester, Michael Denker, Andrew Davison
• Creation of this Collab - Alper Yegenoglu, Johanna Senk
• UNICORE integration - Bernd Schuller
• Task development - Johanna Senk, Bernd Schuller, Alper Yegenoglu, Pietro Quaglio, Michael Denker, Andrew Rowley
• Jupyter Notebooks - Johanna Senk, Bernd Schuller, Alper Yegenoglu, Vahid Rostami
• NEST/PyNN version of microcircuit - Johanna Senk, Sacha van Albada
• ViSTA integration - Benjamin Weyers, Daniel Zielasko
• Support from Collaboratory developers - Yury Brukau, Olivier Amblet, Jean-Denis Courcol and others
• Implementation of the NEST-Spinnaker validation workflow - Robin Gutzen
• Development of the NetworkUnit framework for validation tests - Robin Gutzen, Michael von Papen

References

• Senk J. et al. (2017) A Collaborative Simulation-Analysis Workflow for Computational Neuroscience Using HPC. In: Di Napoli E., Hermanns MA., Iliev H., Lintermann A., Peyser A. (eds) High-Performance Scientific Computing. JHPCS 2016. Lecture Notes in Computer Science, vol 10164. Springer, Cham, doi:https://doi.org/10.1007/978-3-319-53862-4_21

• van Albada SJ, Rowley AG, Senk J, Hopkins M, Schmidt M, Stokes AB, Lester DR, Diesmann M, Furber SB. Performance comparison of the digital neuromorphic hardware SpiNNaker and the neural network simulation software NEST for a full-scale cortical microcircuit model. Front Neurosci (accepted).