Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Most of conventional associative memory models suppose that patterns to be stored are distributed over a pattern space such that each pattern is located at almost the same distance from all the others. In other words, things (concepts) are assumed to be encoded into patterns which do not preserve similarities (correlations) among them. However, this assumption seems implausible, since our actions (e.g. recognition, thinking, and learning) are often based on similarities of concepts. This paper supposes that concepts are represented as hierarchically correlated patterns in which concepts of the same category are encoded into similar patterns, and then proposes a model which successively retrieves patterns belonging to the same category.
