Adaptation of Hopfield Associative Memory Parameters

The paper treats the issue of pattern recognition training in terms of Hopfield associative memory (HAM). The conventional randomization technique is used to determine the exponential extremity of HAM recognition error. The extremity exponent is considered as a function of the training process. In training, the exponent is shown to rise from (1-2p) to (1-2p) where p is the error coefficient at the output of the binary channel of observation (0≤p≤1/2). The HAM capacity grows in the same way.