Computing with Action Potentials

摘要: Most computational engineering based loosely on biology uses continuous variables to represent neural activity. Yet most neurons communicate with action potentials. The view is equivalent using a rate-code for representing information and computing. An increasing number of examples are being discovered in which may not be rate codes. Information can represented the timing potentials, efficiently computed this representation. "analog match" problem odour identification simple solved potential an underlying rhythm. By adapting units effect fundamental change representation problem, we map recognition words (having uniform time-warp) connected speech into same analog match problem. We describe architecture preliminary results such system. Using fast events conjunction rhythm one way overcome limits event-driven computation. When intrinsic hardware much faster than time scale inputs, approach greatly increase effective computation per unit given quantity hardware.