Neuromorphic hardware in action: comparing the implemention of a spiking multivariate classifer model on three neuromorphic platforms
-
1
University Of Sussex, Informatics, United Kingdom
As neuromorphic hardware platforms become more widely available it is important to assess and compare them in the context of the specific task they are required to address. This includes not only adeptness at the task, but also their speed, scalability, power use and ease of implementation.
We report the practical implementation of a bio-inspired, spiking model for multivariate classification on three different platforms: the hybrid digital/analogue Spikey, the digital spike-based SpiNNaker, and GeNN, a meta-compiler for parallel GPU hardware. We assess performance using a benchmark digit classification task.
We found that whilst a different implementation approach was required to create a functioning classifier for each platform, for all three a large fraction of the computing time ended up spent off-device, and on a host machine. Time was spent in combinations of model preparation, encoding suitable input spiking data, shifting data and decoding spike-encoded results. Total power consumed was correspondingly skewed toward the host machine. These observations imply that the efficiency advantage of specialised hardware is easily lost in excessive host-device communication, or non-neuronal parts of the computation. As a result, at smaller model scales, host-only serial implementations outperformed all three platforms.
We conclude that the results emphasise the need to remain “on-device” to leverage the low power consumption and high parallelisation advantages of neuromorphic platforms. In response, we present early results of a revised model which shifts the pre-processing stage of the current model off the host. Instead of employing a host CPU-based algorithm such as neural gas or SOM, we introduce a neuromorphic-based first stage which applies association plasticity and lateral inhibition to self-organise the raw dataset. Classification results are comparable.
Acknowledgements
The research leading to the reported results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 604102 (Human Brain Project). Loan hardware was provided by Manchester University (SpiNNaker) and Heidelberg University (Spikey). M. Schmuker has received funding by a Marie Curie Intra-European Fellowship grant from the European Commission within FP7 (grant no. 331892).
References
Diamond, A. et al., 2014. Classifying chemical sensor data using GPU-accelerated bio-mimetic neuronal networks based on the insect olfactory system. BMC Neuroscience, 15(Suppl 1), p.P77. DOI: 10.1186/1471-2202-15-S1-P77
Furber, S.B. et al., 2013. Overview of the SpiNNaker system architecture. IEEE Transactions on Computers, 62, pp.2454–2467. DOI: 10.1109/TC.2012.142
Khan, M.M. et al., 2008. SpiNNaker: Mapping neural networks onto a massively-parallel chip multiprocessor. In Proceedings of the International Joint Conference on Neural Networks. pp. 2849–2856. DOI: 10.1109/IJCNN.2008.4634199
Nowotny, T., 2011. Flexible neuronal network simulation framework using code generation for NVidia® CUDATM. BMC Neuroscience, 12(Suppl 1), p.P239. DOI: 10.1186/1471-2202-12-S1-P239
Schmuker, M., Pfeil, T. & Nawrot, M., 2014. A neuromorphic network for generic multivariate data classification. Proceedings of the National Academy of Sciences, pp.1–6.
DOI: 10.1073/pnas.1303053111
Diamond A., Nowotny T. & Schmuker M. 2016. Comparing Neuromorphic Solutions in Action: Implementing a Bio-Inspired Solution to a Benchmark Classification Task on Three Parallel-Computing Platforms. Front Neurosci. 8;9:491. doi: 10.3389/fnins.2015.00491.
Keywords:
neuromorphic hardware,
Benchmarking,
bioinspired,
spiking neural networks,
Classification,
self-organisation
Conference:
Neuroinformatics 2016, Reading, United Kingdom, 3 Sep - 4 Sep, 2016.
Presentation Type:
Poster
Topic:
Neuromorphic engineering
Citation:
Diamond
A,
Schmuker
M and
Nowotny
T
(2016). Neuromorphic hardware in action: comparing the implemention of a spiking multivariate classifer model on three neuromorphic platforms
.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2016.
doi: 10.3389/conf.fninf.2016.20.00093
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
14 Jul 2016;
Published Online:
01 Sep 2016.
*
Correspondence:
Dr. Alan Diamond, University Of Sussex, Informatics, Brighton, United Kingdom, A.Diamond@sussex.ac.uk