ASSOCIATIVE MEMORY IN DILUTE RANDOM NETWORKS OF AUTOMATA: OPTIMIZATION AND ADAPTATION IN THE PRESENCE OF NOISE

摘要

We consider random networks of Boolean automata which operate as neural networks for associative memory. Specifically, we consider Boolean automata emulating neurons with synaptic connections and non-linear thresholding dynamics. Techniques for determining the attractor behaviour, basin size and storage capacity are illustrated for a number of synaptic prescriptions. In particular, we consider networks which minimize pattern errors in the presence of a fixed training noise level, and demonstrate how to construct networks which optimally adapt their performances in the presence of retrieval noise.