Variable-Sized Kohonen Feature Map Probabilistic Associative Memory for Sequential Patterns

Itsuki Kitano and Yuko Osana

Keywords

SelfOrganizing Map (Kohonen Feature Map), Probabilistic Association, Successive Learning, Sequential Patterns

Abstract

This paper presents a Variable-sized Kohonen Feature Map Probabilistic Associative Memory for Sequential Patterns (VKFMPAM-SP) which can realize successive (additional) learning for sequential patterns. In the proposed model, since the connection weights of the neurons which are centers of the area corresponding to the stored data are fixed, new patterns can be memorized without destroying memory of the stored patterns. Moreover, in this model, when input pattern is regarded as a new (unknown) pattern and there is not enough neurons which can be used as the area corresponding the input pattern, some neurons can be added in the map layer if needed. And, probabilistic association based on weights distribution (the number of neurons in the area corresponding to each stored pattern) can be realized.

Important Links:



Go Back