Robust Spiking Neural Networks Against Adversarial Attacks

2026-02-24

AI Summary

This work addresses the poor robustness of directly trained spiking neural networks (SNNs) under adversarial attacks, which primarily stems from neurons with membrane potentials near the firing threshold being highly susceptible to minor perturbations that trigger state flips. To mitigate this vulnerability, the authors propose Threshold-Guard Optimization (TGO), which integrates two key components: a potential-aware regularization term in the loss function to encourage membrane potentials to stay away from the firing threshold, and stochastic spiking neurons that replace deterministic firing with a probabilistic mechanism. This study is the first to identify threshold-proximal neurons as a critical robustness bottleneck in SNNs and demonstrates that the synergistic design of potential constraints and probabilistic spiking significantly enhances adversarial resilience. Experiments show that TGO substantially reduces both attack success rates and neuron state-flip probabilities under standard adversarial settings.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Spiking Neural Networks (SNNs) represent a promising paradigm for energy-efficient neuromorphic computing due to their bio-plausible and spike-driven characteristics. However, the robustness of SNNs in complex adversarial environments remains significantly constrained. In this study, we theoretically demonstrate that those threshold-neighboring spiking neurons are the key factors limiting the robustness of directly trained SNNs. We find that these neurons set the upper limits for the maximum potential strength of adversarial attacks and are prone to state-flipping under minor disturbances. To address this challenge, we propose a Threshold Guarding Optimization (TGO) method, which comprises two key aspects. First, we incorporate additional constraints into the loss function to move neurons'membrane potentials away from their thresholds. It increases SNNs'gradient sparsity, thereby reducing the theoretical upper bound of adversarial attacks. Second, we introduce noisy spiking neurons to transition the neuronal firing mechanism from deterministic to probabilistic, decreasing their state-flipping probability due to minor disturbances. Extensive experiments conducted in standard adversarial scenarios prove that our method significantly enhances the robustness of directly trained SNNs. These findings pave the way for advancing more reliable and secure neuromorphic computing in real-world applications.

Problem

• Spiking Neural Networks
• Adversarial Attacks
• Robustness
• Threshold-neighboring Neurons
• State-flipping

Innovation

• Spiking Neural Networks
• Adversarial Robustness
• Threshold Guarding Optimization
• Membrane Potential
• Probabilistic Firing