Quantization of Spiking Neural Networks Beyond Accuracy

arXiv cs.LG / 4/17/2026

📰 NewsDeveloper Stack & InfrastructureIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper argues that quantizing spiking neural networks (SNNs) should be evaluated not only by accuracy, but also by whether the quantized model preserves the firing behavior of its full-precision counterpart.
It shows that quantization choices like method, clipping range, and bit-width can cause significant shifts in firing distributions even when accuracy remains unchanged.
The authors propose using Earth Mover’s Distance (EMD) as a diagnostic metric to measure divergence between firing distributions, making behavior drift visible to deployment-relevant evaluation.
Experiments on SEW-ResNet for CIFAR-10 and CIFAR-100 indicate that uniform quantization can induce distributional drift, whereas LQ-Net-style learned quantization better preserves firing behavior.
The study recommends adding “behavior preservation” as an evaluation criterion alongside accuracy, with EMD as a principled way to assess it for efficient event-driven hardware deployment.

Abstract

Quantization is a natural complement to the sparse, event-driven computation of Spiking Neural Networks, reducing memory bandwidth and arithmetic cost for deployment on resource-constrained hardware. However, existing SNN quantization evaluation focuses almost exclusively on accuracy, overlooking whether a quantized network preserves the firing behavior of its full-precision counterpart. We demonstrate that quantization method, clipping range, and bit-width can produce substantially different firing distributions at equivalent accuracy, differences invisible to standard metrics but relevant to deployment, where firing activity governs effective sparsity, state storage, and event-processing load. To capture this gap, we propose Earth Mover's Distance as a diagnostic metric for firing distribution divergence, and apply it systematically across weight and membrane quantization on SEW-ResNet architectures trained on CIFAR-10 and CIFAR-100. We find that uniform quantization induces distributional drift even when accuracy is preserved, while LQ-Net style learned quantization maintains firing behavior close to the full-precision baseline. Our results suggest that behavior preservation should be treated as an evaluation criterion alongside accuracy, and that EMD provides a principled tool for assessing it.