Learning system for neurocircuit network

Abstract

Abstract PURPOSE:To prepare multiple pseudo learning patterns and to attain an effect near pattern recognition by means of a multi-template method by permitting a neurocircuit network to execute learning by means of the multi-template method in a learning device, for example. CONSTITUTION:A pseudo learning pattern generation part 100 reads the learning patterns from a learning pattern storage part 72 and transmits them to a category decision part 101. It sets the pattern from the pattern generation part 100 as an unknown input pattern and decides the category by a 'k-vicinity method' by setting the learning pattern of the storage part 72 as a dictionary. In the 'k-vicinity method', a distance between all dictionary patterns and the input pattern is calculated, k-number of the patterns of small distance are selected for respective categories and the average of k-number of distances is set to be the distance of the category and the input pattern. Thus, the pattern nearest to the input pattern is set to be a decision result.

Inventors

• 👤 Akihiro Kimura
• 👤 晋太 木村

Bibliographic Details