From a8767bd91210040d968e5998a51c86d1be529493 Mon Sep 17 00:00:00 2001 From: Alessandro Pierro Date: Fri, 20 Oct 2023 19:36:43 +0200 Subject: [PATCH] Add draft --- paper.bib | 21 +++++++++++++++++++++ paper.md | 27 ++++++++++++++++----------- 2 files changed, 37 insertions(+), 11 deletions(-) diff --git a/paper.bib b/paper.bib index 833924b5..0aae5b2a 100644 --- a/paper.bib +++ b/paper.bib @@ -14,4 +14,25 @@ @inproceedings{lavabo keywords = {bayesian optimization, neuromorphic computing, asynchronous computing}, location = {Santa Fe, NM, USA}, series = {ICONS '23} +} + +@ARTICLE{10.3389/fnins.2018.00816, + +AUTHOR={Rhodes, Oliver and Bogdan, PetruĊ£ A. and Brenninkmeijer, Christian and Davidson, Simon and Fellows, Donal and Gait, Andrew and Lester, David R. and Mikaitis, Mantas and Plana, Luis A. and Rowley, Andrew G. D. and Stokes, Alan B. and Furber, Steve B.}, + +TITLE={sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker}, + +JOURNAL={Frontiers in Neuroscience}, + +VOLUME={12}, + +YEAR={2018}, + +URL={https://www.frontiersin.org/articles/10.3389/fnins.2018.00816}, + +DOI={10.3389/fnins.2018.00816}, + +ISSN={1662-453X}, + +ABSTRACT={This work presents sPyNNaker 4.0.0, the latest version of the software package for simulating PyNN-defined spiking neural networks (SNNs) on the SpiNNaker neuromorphic platform. Operations underpinning realtime SNN execution are presented, including an event-based operating system facilitating efficient time-driven neuron state updates and pipelined event-driven spike processing. Preprocessing, realtime execution, and neuron/synapse model implementations are discussed, all in the context of a simple example SNN. Simulation results are demonstrated, together with performance profiling providing insights into how software interacts with the underlying hardware to achieve realtime execution. System performance is shown to be within a factor of 2 of the original design target of 10,000 synaptic events per millisecond, however SNN topology is shown to influence performance considerably. A cost model is therefore developed characterizing the effect of network connectivity and SNN partitioning. This model enables users to estimate SNN simulation performance, allows the SpiNNaker team to make predictions on the impact of performance improvements, and helps demonstrate the continued potential of the SpiNNaker neuromorphic hardware.} } \ No newline at end of file diff --git a/paper.md b/paper.md index 3ce0a8f4..5f4eccd6 100644 --- a/paper.md +++ b/paper.md @@ -52,7 +52,6 @@ As this short list shows, JOSS papers are only expected to contain a limited set # Summary - - Challenges of optimization and opportunities of neuromorphic computing - Scalability, low latency, optimality, energy - Neuromorphic computing provides fine-grained parallel and event-driven copmutation, low energy consumption @@ -62,27 +61,33 @@ As this short list shows, JOSS papers are only expected to contain a limited set - Supports the community in developing algorithms that are iterative, discrete, and distributed - We leveraged the library architecture to develop multi-backend QUBO and QP solvers, and received contirbutions from the community for a Bayesian `[@lavabo]` and LCA solvers - # Statement of need A Statement of need section that clearly illustrates the research purpose of the software and places it in the context of related work. -- Energy Delay Product +- Difficulty of programmability of neuromorphic + - SpynNaker: more suited for neuroscience, hard to use to program algorithms + - NENGO, NEST, BrainScale, brian + - Previous implementations of neuromorhic optimization algorithms were mostly standalone solvers, with primitive user experience and exposing low-level implementation details to the user-facing APIs -# Acknowledgements +- Lava offers: -We acknowledge contributions from Chinonso Onah, Gavin Parpart, and Shay Snyder. +- Lava Optimization introduce a clear separation of abstraction layers, modularity/composability/ortogonality/DRY, and easy to change +- Lava Optim: on-chip online tracking of best ongoing solutions instead of post-processing recorded activity (improve performance, reduce memory/bandwidth requirements) +- Lava Optimization includes the mathematical primitives needed for optimization problems/algorithms +- Solver tuner / visualization / reporting / profiling analysis tools specifically for optimization +- Easy to extend / possible future directions: regularization terms and higher-order optimization +- Lava Optimization also includes applications specific solvers / interfaces (scheduling, tsp, clustering, vrp) +- Modern software practices, for DevOps / automated testing, distributed with pip -# [OPTIONAL] On-going research using the library +- Mention (if applicable) a representative set of past or ongoing research projects using the software and recent scholarly publications enabled by it.: lava BO and lava MPC -Mention (if applicable) a representative set of past or ongoing research projects using the software and recent scholarly publications enabled by it. +# Acknowledgements -# Financial support +We acknowledge contributions from Chinonso Onah, Gavin Parpart, and Shay Snyder. -Acknowledgement of any financial support. +# Financial support This project was developed by the Neuromorphic Computing Lab at Intel Corporation. # References - -A list of key references, including to other software addressing related needs. Note that the references should include full names of venues, e.g., journals and conferences, not abbreviations only understood in the context of a specific discipline. \ No newline at end of file