Oscillatron-1: a connectionist unified theory of cognition

A novel artificial neural network architecture is introduced that implements declarative as well as procedural memory. This network contains an internal clock of central pattern generators that oscillates in synchrony with feature detector neurons. These properties allow the network to associatively store and replay moving sensory images by storing a time stamp with each snapshot within high-order correlation synapses. When combined with conventional neural procedural memory and the dynamics of a blackboard system, this architecture should be able to recognize and mentally manipulate images and their abstractions and to remember its own thought processes. The first results from simulations demonstrate that it can photographically memorize and replay moving images of single lines when given visual clues about one snapshot from the sequence.<>