A Pattern Machine

This page introduces an idea for a possible mode of operation of a recurrent spiking neural network. The aim is to get the network to be able to store, recall and recognize complex and precise patterns, such as a music score, the London tube map, or a paragraph of text.

It can be useful to view the functions of such a network, at an abstract "pattern machine" level rather than in terms of individual neuron or synapse activations. In order to do this we need to make a number of assumptions that the network is capable of implementing an auto-associative memory. We also need a small amount of control circuitry to gate learning and recall.

This site presents a collection of assumptions, along with a high level simulator of the dynamics of such patterns which can perform basic replay, memory, chunking, and recognition of patterns such as MIDI files, words from letters, and the London Underground map.

www.neurogems.org