Neuromorphic Selective Attention Systems

Abstract

Selective attention mechanisms allow sensory systems with limited processing capacity to function in real time, independently of the size of the stimulus input space. They can be particularly useful in artificial vision systems, where the amount of information provided by the sensors typically exceeds the system's processing capacity. We describe a very large-scale integration (VLSI) neuromorphic device and a spike-based communication infrastructure for constructing multichip vision systems based on models of selective attention. These hardware models reproduce many of the behaviors of biological selective attention systems. We demonstrate the feasibility of this approach by showing experimental results from a hardware two-chip vision system that performs active tracking and selective scanning tasks.