Computer Architecture Group
eFutures Workshop December 2013
Code Generation for Neural Computation - High-level descriptions to efficient simulations
Overview
Michael Hull and Simon Moore ran a workshop on the 17/18th Dec 2013 at Trinity College, Cambridge, with the aim of bringing together people to discuss the role of code-generation in large scale neuroscience simulations.
Code generation is an exciting and active area of research in neuroscience tool development, allowing scientists to turn their ideas into simulation results quickly. Tools and languages are currently being developed by a number of groups, which allow modellers to define the behaviours of neurons and their interconnections at a high-level which can then be compiled to fast, architecture-specific, machine code.
One particular application for code-generation in simplifying the interface to powerful hardware platforms, ranging from GPGPUs to the BlueHive and SpiNNaker projects, which shields the modeller from the esoteric aspects of development. One aim of the workshop was to bring together people developing or interested in using such tools to exchange experiences and discuss technical difficulties and successes. We were also interested to hear general updates about the development of other tools in computational neuroscience.
Programme
Thursday morning - Project overviews:
- Steve Furber : Toward massively parallel compute for the HBP
- Simon Moore : Massively parallel FPGA compute: from custom computing to vector processing
Thursday morning -- Short lightening talks:
- Boris Marin Template based code generation for networks of hybrid systems
- Mike Vella libNeuroML - NeuroML meets Python
- Joel Chavas
- Robert Horte-Merrison DSvis: A tool for interactively exploring dynamical systems using GPU computing
Thursday mid-morning - Code-generation, mapping to fixed-point & event-based computing issues:
- Thomas Nowotny: The GeNN framework (GPGPU) and the recent development of a BRIAN2 interface
- David Lester: Using ISO fixed point arithmetic for neural modelling: theory and practice
- Mike Hull: Tadpole on FPGA: mapping floating point equations into integers using NineML
- Andrew Brown: Atomic Computing: broadening the simulation infrastructure
- Angus Silver: OpenSource Brain: Making models more transparent and accessible with NeuroML2/LEMS
- Steve Marsh/Rob Cannon: LEMS-lite - An intermediate format for code-generation
- Padraig Gleeson: Code generation in NeuroML/LEMS and the SOM format
Attendees
- Thomas Nowotny (Sussex)
- Steve Furber (Manchester)
- David Lester (Manchester)
- Michael Hopkins (Manchester)
- Sergio Davies (Manchester)
- Andrew Brown (Southampton)
- Kier Dugan (Southampton)
- Rob Mills (Southampton)
- Tom J Kazmierski (Southampton)
- Angus Silver (UCL)
- Boris Marin (UCL)
- Padraig Gleeson (UCL)
- Robert Cannon (Edinburgh)
- Mike Hull (Cambridge)
- Simon Moore (Cambridge)
- Theo Markettos (Cambridge)
- Paul Fox (Cambridge)
- Steve Marsh (Cambridge)
- Matt Naylor (Cambridge)
- Joel Chavas (Paris)
- Bob Merrison-Horte (Plymouth)
- Mike Vella (Cambridge)