Summary: Spiking Sequence Learning using Maximum
Likelihood: Hopfield Networks
David Barber # Felix Agakov
Division of Informatics, University of Edinburgh, Edinburgh EH1 2QL, UK
email@example.com firstname.lastname@example.org http://anc.ed.ac.uk
July 15, 2002
We consider the learning of correlated temporal sequences from a statistical
viewpoint, using Maximum Likelihood to derive a simple local learning rule for
an idealized spiking neuron model. The rule is capable of robustly storing mul
tiple sequences of correlated patterns in a recurrent network model, unlike other
prescriptions based on pattern correlations alone.
Simple models of temporal sequence storage devices are typically based on recurrent net
works that model the hippocampal region CA3 with training sequence inputs represented
# Corresponding author.
by projections from entorhinal cortex and dentate gyrus (Shon et al., 2002). Mechanisms
by which a network can learn to store temporal sequences are usually based on local