TIME-DELAY ART FOR SPATIOTEMPORAL PATTERNS

This paper proposes a new self-organizing neural network for spatio-temporal patterns called Time-Delay ART (TD-ART). The TD-ART is a nearest neighbor classifier that stores arbitrary length of spatio-temporal patterns. It has at least three layers. The Layer-1 and the Layer-2 classify input spatial patterns one by one: the algorithm is based on the Carpenter/Grossberg net algorithm. The Layer-2 and the Layer-3 classify and memorize the sequence of the spatial patterns classified in the Layer-2. The connections between the Layer-2 and the Layer-3 are adjusted both in weight and in time-delay to deal with spatio-temporal patterns. Although the construction and operation of TD-ART are very simple, it has many features such as its self-organization, classification, and memorization abilities for spatio-temporal patterns. In addition, the TD-ART can distinguish different sequences composed of same patterns such as A --> C --> T and C --> A --> T by unsupervised learning, and can deal with many sequences of different lengths.