Encoding Static and Temporal Patterns with a Bidirectional Heteroassociative Memory

[摘要] Brain-inspired, artificial neural network approach offers the ability to develop attractors foreach pattern if feedback connections are allowed. It also exhibits great stability andadaptability with regards to noise and pattern degradation and can perform generalizationtasks. In particular, the Bidirectional Associative Memory (BAM) model has shown greatpromise for pattern recognition for its capacity to be trained using a supervised orunsupervised scheme. This paper describes such a BAM, one that can encode patterns ofreal and binary values, perform multistep pattern recognition of variable-size time series andaccomplish many-to-one associations. Moreover, it will be shown that the BAM can begeneralized to multiple associative memories, and that it can be used to store associationsfrom multiple sources as well. The various behaviors are the result of only topologicalrearrangements, and the same learning and transmission functions are kept constantthroughout the models. Therefore, a consistent architecture is used for different tasks, therebyincreasing its practical appeal and modeling importance. Simulations show the BAM'svarious capacities, by using several types of encoding and recall situations.

[学科分类] 应用数学