Skip to main content.


This pattern machine is intended as an exploration of a way of visualising higher level behaviour of recurrent spiking networks. The underlying machine architecture (variants on multi-state machines) provides a general purpose 'virtual machine' for doing useful things like learning and replaying complex patterns, predicting inputs and recognising novelty, and chunking of long patterns into smaller ones.

In my opinion this general purpose virtual machine could be implemented using real neurons, given what we know about their dynamics. Whether it is or not is hard to answer at present, since we are not yet able to do simultaneous recordings from more than a few 10s of neurons, and it is hard to extrapolate network behaviour from single neurons.

Fred Howell (fred at