SpikeMatch: Semi-Supervised Learning with Temporal Dynamics of Spiking Neural Networks

2025-09-26

AI Summary

Current semi-supervised learning (SSL) methods for spiking neural networks (SNNs) are scarce and prone to confirmation bias due to inadequate modeling of temporal dynamics. Method: We propose SpikeMatch—the first SSL framework explicitly incorporating SNN temporal dynamics. It leverages the leak factor of leaky integrate-and-fire neurons to modulate temporal responses, generating multi-view pseudo-labels for the same input; integrates strong/weak data augmentations with multi-prediction consistency constraints; and enables discriminative learning on weakly labeled samples via single-model co-training. Contribution/Results: SpikeMatch effectively mitigates confirmation bias inherent in conventional pseudo-labeling approaches and significantly improves few-shot generalization. On multiple standard benchmarks, it substantially outperforms existing SNN-adapted SSL methods, empirically validating the critical role of explicit temporal dynamic modeling in SNN semi-supervised learning.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Spiking neural networks (SNNs) have recently been attracting significant attention for their biological plausibility and energy efficiency, but semi-supervised learning (SSL) methods for SNN-based models remain underexplored compared to those for artificial neural networks (ANNs). In this paper, we introduce SpikeMatch, the first SSL framework for SNNs that leverages the temporal dynamics through the leakage factor of SNNs for diverse pseudo-labeling within a co-training framework. By utilizing agreement among multiple predictions from a single SNN, SpikeMatch generates reliable pseudo-labels from weakly-augmented unlabeled samples to train on strongly-augmented ones, effectively mitigating confirmation bias by capturing discriminative features with limited labels. Experiments show that SpikeMatch outperforms existing SSL methods adapted to SNN backbones across various standard benchmarks.

Problem

• Develops semi-supervised learning for spiking neural networks
• Leverages temporal dynamics to generate reliable pseudo-labels
• Mitigates confirmation bias with limited labeled data

Innovation

• Leverages SNN temporal dynamics for pseudo-labeling
• Uses co-training framework with weak and strong augmentations
• Generates reliable labels from single SNN multiple predictions