Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch

2024-10-31

AI Summary

To address training instability, low accuracy, and slow convergence in Time-to-First-Spike (TTFS) spiking neural networks—caused by inherent single-spike sparsity—this paper proposes the first end-to-end trainable TTFS framework. Our method introduces three key innovations: (1) a synergistic parameter initialization and dynamic temporal normalization scheme that significantly enhances training stability; (2) a lightweight temporal output decoding strategy that reduces inference latency; and (3) theoretical analysis and empirical validation establishing the necessity of average pooling for preserving the single-spike property of TTFS. Evaluated on four benchmarks—MNIST, Fashion-MNIST, CIFAR-10, and DVS-CIFAR10—the framework achieves state-of-the-art accuracy (up to 99.48%), accelerates convergence by ~40%, reduces training variance by over 60%, and cuts inference latency by up to 35%.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Learning & Optimization for CV
• Machine Learning: Time-Series/Data Streams

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Spiking Neural Networks (SNNs), with their event-driven and biologically inspired operation, are well-suited for energy-efficient neuromorphic hardware. Neural coding, critical to SNNs, determines how information is represented via spikes. Time-to-First-Spike (TTFS) coding, which uses a single spike per neuron, offers extreme sparsity and energy efficiency but suffers from unstable training and low accuracy due to its sparse firing. To address these challenges, we propose a training framework incorporating parameter initialization, training normalization, temporal output decoding, and pooling layer re-evaluation. The proposed parameter initialization and training normalization mitigate signal diminishing and gradient vanishing to stabilize training. The output decoding method aggregates temporal spikes to encourage earlier firing, thereby reducing the latency. The re-evaluation of the pooling layer indicates that average-pooling keeps the single-spike characteristic and that max-pooling should be avoided. Experiments show the framework stabilizes and accelerates training, reduces latency, and achieves state-of-the-art accuracy for TTFS SNNs on MNIST (99.48%), Fashion-MNIST (92.90%), CIFAR10 (90.56%), and DVS Gesture (95.83%).

Problem

• Stabilizes training of Time-to-First-Spike Spiking Neural Networks
• Reduces latency in TTFS SNNs through temporal output decoding
• Improves accuracy of TTFS SNNs on multiple datasets

Innovation

• Parameter initialization stabilizes TTFS SNN training
• Temporal output decoding reduces spike latency
• Re-evaluated pooling layers enhance single-spike efficiency