PTQ4SNN: Membrane-Aware Post-Training Quantization for Spiking Neural Networks

2026-08-07

AI Summary

This work addresses the challenge of quantizing spiking neural networks (SNNs) under low-bit constraints, where membrane potentials are typically represented in floating-point format, leading to complex distributions and high sensitivity to threshold perturbations that hinder effective quantization and induce error accumulation. To overcome this, the authors propose a post-training quantization framework that jointly quantizes both weights and recurrent membrane potential states without requiring retraining, applicable to both convolutional SNNs and spiking-driven Transformers. The method introduces a channel-wise uniform scaling bridge to align the scales of membrane potentials and weights and employs a mixed-precision allocation strategy based on neuronal firing activity and quantization sensitivity, optimizing accuracy under an average bit-width budget. Experiments demonstrate that with weights quantized to 4 bits and membrane potentials to approximately 4 bits, the models maintain high accuracy on image classification and semantic segmentation tasks.

Abstract

Spiking neural networks (SNNs) enable sparse and event-driven computation, but their low-bit deployment remains incomplete because recurrent membrane states are commonly retained in floating point even after weight quantization. Quantizing these states is challenging because their distributions differ across channels and from the preceding weights, while small perturbations near the firing threshold may alter spike decisions and accumulate over time. We propose PTQ4SNN, a membrane-aware post-training quantization framework that jointly quantizes weights and recurrent membrane states using only a small calibration set. First, a channel-wise Unified Scale Bridge constrains the membrane scale as s_mem,c = s_w,c * 2^k_c, adapting to membrane distributions while enabling shift-compatible scale conversion. Second, Mixed-Precision Bit Allocation assigns 2/4/8-bit precision to membrane channels according to firing activity and quantization sensitivity under an average-bit budget. The framework operates on reusable projection-LIF pairs and supports both convolutional SNNs and spike-driven Transformers without backbone retraining. Experiments on static and event-based classification and semantic segmentation show that PTQ4SNN effectively preserves model accuracy under W4 quantization and approximately 4-bit membrane precision.

Problem

• Spiking Neural Networks
• Post-Training Quantization
• Membrane Potential
• Low-bit Deployment
• Quantization Sensitivity

Innovation

• post-training quantization
• spiking neural networks
• membrane potential quantization
• mixed-precision allocation
• event-driven computation