Energy-Aware Spike Budgeting for Continual Learning in Spiking Neural Networks for Neuromorphic Vision

Researchers developed an energy-aware spike budgeting framework for spiking neural networks (SNNs) that dramatically reduces power consumption while improving accuracy for continual learning. The method cuts spike rates by up to 47% on frame-based datasets like CIFAR-10 and boosts accuracy by 17.45 percentage points on event-based datasets like DVS-Gesture. By integrating experience replay with adaptive spike scheduling, it enables ultra-low-power perception for evolving environments without catastrophic forgetting.