What can neuromorphic event-driven precise timing add to spike-based pattern recognition?

摘要: This letter introduces a study to precisely measure what an increase in spike timing precision can add spike-driven pattern recognition algorithms. The concept of generating spikes from images by converting gray levels into timings is currently at the basis almost every spike-based modeling biological visual systems. use naturally leads incorrect artificial and redundant and, more important, also contradicts findings indicating that processing massively parallel, asynchronous with high temporal resolution. A new for acquiring information through pixel-individual level-crossing sampling has been proposed recent generation neuromorphic sensors. Unlike conventional cameras, these sensors acquire data not fixed points time entire array but amplitude changes their input, resulting optimally sparse space time-pixel individually timed only if new, previously unknown available event based. uses resolution spiking output event-based show lowering degrades performance on several tasks specifically when reaching range machine vision acquisition frequencies 30-60i¾ Hz. theory characterize separability between classes each shows provides up 70% generated frame-based as used standard vision, thus drastically increasing objects. Experiments real amount loss correlated precision. Our information-theoretic highlights potentials both practical applications theoretical investigations. Moreover, it suggests representing precise sequence times reported retina offers considerable advantages neuro-inspired computations.