Improving pattern reconstruction in neural networks by activity dynamics
Department of Physics,
The Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong
Accepted: 26 October 1996
I study the averaged dynamical behaviour of neural networks over an extended monitoring period, and consider pattern reconstruction procedures by activity clipping, selectively freezing, or sequentially freezing the dynamic nodes. They enable the retrieval precision to be improved, the basin of attraction to be widened, or the storage capacity to be increased, even when the information is not efficiently embedded in the synaptic weights.
PACS: 87.10.+e – General, theoretical, and mathematical biophysics (including logic of biosystems, quantum biology, and relevant aspects of thermodynamics, information theory, cybernetics, and bionics) / 05.20.-y – Statistical mechanics / 02.50.-r – Probability theory, stochastic processes, and statistics
© EDP Sciences, 1996