Spiking neural networks get better at handling event disruptions
PAFS: Perturbation-Aware Firing Scaffolds for Robust Event-Based SNNs
Neural and Evolutionary Computing
Summary
Spiking neural networks use precise timing to process information, but small changes like event loss or timing errors can make them less accurate. The authors introduce a way to correct for these problems after the network is trained, by looking at the network’s later activity patterns to adjust predictions. Their method, called PAFS, improves how well these networks perform in challenging conditions without retraining the whole system. This helps make event-based AI systems more reliable when inputs are imperfect.
What this means in practice
- •For autonomous vehicle engineers: Use PAFS to enhance event-based vision systems’ robustness against sensor noise and timing errors during driving.
- •For robotics system developers: Integrate PAFS calibration to improve real-time decision making in robots using spiking neural networks under variable input conditions.
Authors
Ning Liu, Jun Sun, Hao Guo, Sen Shen, Xiangfei Yang, Jie Yang, Lan Wu, Shangke Lyu, Thomas Bräunl
Abstract
Event-based spiking neural networks (SNNs) encode information through temporal firing dynamics, but the same stateful computation makes their predictions sensitive to event loss, timestamp jitter, and speed variation. These perturbations alter evidence accumulation, yet late-layer firing trajectories can retain error-diagnostic structure that is absent from the final logits alone. We introduce Perturbation-Aware Firing Scaffolds (PAFS), a labeled post-training calibration method that exploits this structure without updating the pretrained backbone. PAFS associates standardized late-layer leaky integrate-and-fire trajectories with the class-probability residuals of an Adapter anchor, then applies singular value decomposition to extract compact residual-aligned channel directions. At inference, firing trajectories from the same backbone pass are projected onto this fixed low-rank scaffold and combined with anchor statistics to predict a residual logit correction. Across four matched dataset-backbone settings spanning three event datasets and two spiking Transformer architectures, PAFS improves corrupted-average and worst-condition accuracy over the corresponding Adapter anchors, with maximum gains of 3.06 and 3.65 percentage points. In the DailyAction-DVS/Spikingformer setting, the added trainable parameters amount to approximately 0.34% of the frozen backbone.