Information processing in networks with multiple quasi-stable attractor states
Paul Miller
Department of Biology
Brandeis University
Clusters of neurons with strong excitatory connections to each other can lead to networks with multiple activity states, each state defined by the set of clusters with stably high versus low neural firing rates. Noise or adaptive processes can render the states quasi-stable (for example as in a relaxation oscillator) such that in the presence of constant input, or even more so, following successive inputs, the network can progress through a sequence of different activity states. There is significant evidence for such state sequences in neural dynamics, and here we study relatively simple models, assessing the extent to which such state sequences might subserve information processing and behavior in a range of cognitive tasks.