Quantization of Spiking Neural Networks Beyond Accuracy

2026-04-15

AI Summary

This work addresses a critical gap in the evaluation of quantized spiking neural networks (SNNs), which has predominantly focused on accuracy while overlooking the impact of quantization on neuronal spiking behavior—a key determinant of sparsity and computational load during deployment. To remedy this, the study introduces spiking behavior fidelity as an essential evaluation criterion and proposes using Earth Mover’s Distance (EMD) to quantify the discrepancy between pre- and post-quantization spiking distributions. Experiments on SEW-ResNet over CIFAR-10 and CIFAR-100 demonstrate that uniform quantization, despite preserving high accuracy, induces substantial distributional shifts, whereas learning-based quantization methods such as LQ-Net better retain the original spiking characteristics. These findings underscore that behavioral consistency should be regarded as an evaluation metric independent of accuracy in SNN quantization.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

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.

Problem

• Spiking Neural Networks
• Quantization
• Firing Behavior
• Accuracy
• Deployment

Innovation

• Spiking Neural Networks
• Quantization
• Firing Behavior Preservation
• Earth Mover's Distance
• Learned Quantization