Reproduce “Efficient Deployment of Spiking Neural Networks on SpiNNaker2 for DVS Gesture Recognition Using Neuromorphic Intermediate Representation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Sirine Arfa, Bernhard Vogginger, Chen Liu, Johannes Partzsch, Mark Schone, Christian Mayr

Researchers developed an end-to-end pipeline for deploying deep Spiking Neural Networks (SNNs) on the SpiNNaker2 neuromorphic chip, achieving 94.13% on-chip accuracy for DVS gesture recognition with an 8-bit Quantization Aware Training (QAT) approach and reducing model memory footprint by 75%. The work demonstrates SpiNNaker2's capability for energy-efficient, high-accuracy SNN inference while enhancing interoperability through the Neuromorphic Intermediate Representation (NIR).

Key insights

  • Quantization Aware Training (QAT) with adaptive threshold scaling is more effective for accurate and memory-efficient SNN inference on neuromorphic processors like SpiNNaker2, significantly outperforming Post Training Quantization (PTQ).
  • Quantizing both synaptic weights and neuron firing thresholds is crucial for maximizing memory savings in SNNs while minimizing accuracy degradation, a key aspect often overlooked in prior work.
  • SpiNNaker2 demonstrates its potential as a high-performing and energy-efficient platform for event-driven SNN inference tasks, achieving competitive accuracy for DVS gesture recognition with minimal energy consumption.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.