Programmable Silicon Retina on Pixel Processor Array
Researchers have developed a multi-stage Silicon Retina model on the SCAMP-5 Pixel Processor Array, incorporating spatial filtering and gain control inspired by biological retinas. This bio-inspired model demonstrated a 13% reduction in saliency prediction loss compared to standard dynamic vision sensors, while also decreasing the event rate by approximately 47%. The findings suggest that this "information distillation" mechanism can create more efficient representations for downstream neural networks, particularly beneficial for bandwidth-constrained edge applications. AI
IMPACT This bio-inspired model could lead to more efficient AI processing on edge devices by distilling information.