A Biologically Plausible Neural Network Able to Learn Complex Behaviors

摘要

This paper advances work on a simple, modular network with behaviors analogous to planning, refabrication of memory, and classical, operant, and backward conditioning. The network recalls events and actions serially, distinguishes between real and recalled events, is context sensitive, and is able to respond to the absence as well as the presence of a stimulus. Some of the mechanisms seem to represent self-awareness and imagination. Accordingly, the results suggest that consciousness may be achieved by surprisingly simple networks. The current work discusses biologically plausible mechanisms, not previously modeled, by which the network can create representations of "sensory" events.