BSO: Binary Spiking Online Optimization

Binary Spiking Neural Networks (BSNNs) offer promising efficiency advantages for resource-constrained computing. However, their training algorithms often require substantial memory overhead due to latent weight storage and temporal processing requirements. To address this, we propose Binary Spiking Online Optimization (BSO), a novel online training algorithm that significantly reduces training memory while maintaining energy efficiency during the forward pass. BSO achieves this through two key innovations: memory requirements independent of time steps and elimination of latent weight storage. BSO directly updates binary weights through flip signals using the online training framework. These signals are triggered when gradient momentum exceeds a threshold, eliminating the need for latent weight during training. To leverage the inherent temporal dynamics of BSNNs, we further introduce T-BSO, a temporal-aware variant that captures gradient information across time steps for adaptive threshold adjustment. Theoretical analysis establishes convergence guarantees for both BSO and T-BSO, with formal regret bounds characterizing their convergence rates. Extensive experiments demonstrate that both BSO and T-BSO achieve superior optimization performance compared to existing training methods for binary spiking neural networks while significantly reducing memory overhead during training. Our work presents the first effective integration of online training with BSNNs, advancing their practical applicability in resource-constrained scenarios.