Self-Organizing Map-based Probabilistic Associative Memory for Sequential Patterns

In this paper, we propose a Self-Organizing Map-based Probabilistic Associative Memory for Sequential Patterns (SOMPAM-SP). The proposed model is based on Self-Organizing Map and it has an Input/Output Layer and a Map Layer. The Input/Output Layer is divided into three parts; (1) Input Part, (2) Output Part and (3) Brief Degree Part. In this model, stored pattern sequences are divided into pattern sets composed of two patterns (the pattern at the time t and the pattern at the time t + 1) and each set is memorized with its own brief degree. In this model, probabilistic associations based on brief degree for sequential binary/analog patterns including common term(s) can be realized. Moreover, it can also realize additional learning by adding new neuron if needed. We carried out a series of computer experiments and confirmed that the proposed model can realize probabilistic associations of sequential patterns and additional learning.