Towards an Expressivity-Normalized Energy-Demand Comparison of ANNs and SNNs

2026-08-30

AI Summary

研究通过构建分析框架，对比了ANNs与SNNs在处理时间序列数据时的能效比，考虑了表达能力匹配下的网络架构和数据特性的影响。

Abstract

Spiking neural networks (SNNs) are often regarded as energy-efficient alternatives to artificial neural networks (ANNs), yet their advantage depends critically on both network architecture and data properties. We develop an analytical framework to compare fully-connected ReLU ANNs and integrate-and-fire SNNs for time-series data with respect to their theoretical energy efficiency at matched expressive capacity. By relating an inference-energy model to theoretical bounds on representational expressivity, we derive an expressivity-normalized efficiency ratio and explicit thresholds in network width, spike sparsity, and ANN depth scaling. Our analysis characterizes the regimes in which event-driven computation offsets the temporal overhead of SNNs, providing capacity-aware principles for designing energy-efficient temporal networks. It shows that ANNs exceed SNNs in expressivity-normalized efficiency only in specific regimes.

Problem

• Spiking Neural Networks
• Artificial Neural Networks
• Energy Efficiency
• Expressive Capacity
• Time-series Data

Innovation

• expressivity-normalized efficiency
• spike sparsity
• event-driven computation