Improving bidirectional associative memory performance by unlearning

Abstract

An encoding strategy for improving the noise tolerance and capacity of Kosko's bidirectional associative memory is proposed. Energy " a corresponding to pattem pairs that are to be stored are enhanced and simultaneously, unwanted or spurious states are eliminated. The method is an extension of the multiple training procedure that has been described in the literature for inducing local minima at desired locations. An additional unlearning term in the energy expression is included to eliminate spurious states. Spurious states and parameter values for constructing the network are determined experimentally. Computer simulations show that unlearning increases the network's storage capacity and its tolerance to noise to a level not achievable by multiple training alone.

Improving Leung's bidirectional learning rule for associative memories

“…4], for general information about associative memories, respectively, networks). During the years, various different strategies were proposed in order to design such networks (without claim of completeness we mention Grossberg [3], Anderson [1], Kohonen [8], Cohen and Grossberg [2], Kosko [10], [11], Hassoun [4], Wang et al [19], [20], Srinivasan and Chia [18], Shanmukh and Venkatesh [16], Zhang et al [22], Leung [14], [15], Wang et al [21], [23], and, finally, Sommer and Palm [17]) which proved to be suitable in different settings. In the following, we are especially interested in the idea proposed by Leung [15] classical perceptron learning rule for BAMs and proved a kind of optimality property for this new learning strategy (which he called bidirectional learning).…”