AI RESEARCH
Vision SmolMamba: Spike-Guided Token Pruning for Energy-Efficient Spiking State-Space Vision Models
arXiv CS.CV
•
ArXi:2604.25570v1 Announce Type: new Spiking Transformers have shown strong potential for long-range visual modeling through spike-driven self-attention. However, their quadratic token interactions remain fundamentally misaligned with the sparse and event-driven nature of spiking neural computation. To address this limitation, we propose Vision SmolMamba, an energy-efficient spiking state-space architecture that integrates spike-driven dynamics with linear-time selective recurrence.