Two coding strategies for bidirectional associative memory

Abstract

Enhancements of the encoding strategy of a discrete bidirectional associative memory (BAM) reported by B. Kosko (1987) are presented. There are two major concepts in this work: multiple training, which can be guaranteed to achieve recall of a single trained pair under suitable initial conditions of data, and dummy augmentation, which can be guaranteed to achieve recall of all trained pairs if attaching dummy data to the training pairs is allowable. In representative computer simulations, multiple training has been shown to lead to an improvement over the original Kosko strategy for recall of multiple pairs as well. A sufficient condition for a correlation matrix to make the energies of the training pairs be local minima is discussed. The use of multiple training and dummy augmentation concepts are illustrated, and theorems underlying the results are presented.

The Trainable Model of Memorization of Movement Sequence Based on Heteroassociative Neural Network

“…Между тем, в последние два десятилетия возникла теория гетероассоциативных сетей, в которых по одному из объектов пары восстанавливается другой объект [5]. Появились и более сложные алгоритмы обучения таких сетей заучиванию, которые, однако, не применялись при изучении памяти человека [6][7][8][9]. Интересно отметить, что принцип работы сетей этого типа изложен и в художественной литературе [10].…”

Using Multi-layered Feed-forward Neural Network (MLFNN) Architecture as Bidirectional Associative Memory (BAM) for Function Approximation

“…In [15], a new bidirectional hetero-associative memory is defined which encompasses correlational, competitive and topological properties and capable of increasing its clustering capability. Other related work can be found in [16], [17], [18] and [19] in which either by adding dummy neuron, increasing in number of layers or manipulating the interconnection among neurons in each layer, the issue of performance improvement of BAM is addressed. Even some new learning algorithms were introduced to improve the performance of original BAM and can be found in detail in [20]- [24].…”

New Strategies for Associative Memories

“…Figure 4 illustrate the structure of MBAM. While kosko expanded the unidirectional auto-associative to bidirectional associative processes [9], by utilizing the correlation matrix given in equation (1), the system able to retrieve the nearest pair given any pair (X,Y), where (X) is the input pattern and (Y) is the output pattern with the help of multidirectional coding and encoding process. However, sometimes the encoding could not ensure that the saved pairs are at local minimum and for this reason results in incorrect recall.…”

“…Wang's generalized correlation matrix is [9]: M= ∑ × × (4) Where (qi) is showed as the weight of( × ) and it is a positive real number. It indicates the minimum number of times for utilizing a pattern pair (xi , yi) for training to ensure correct recall of that pair , but it also had a problem , it guarantee to recall only one training pair.…”