Hiroshi Tsukimoto

The functional localization of neural networks using genetic algorithms

We presented an algorithm for extracting Boolean functions (propositions, rules) from the units in trained neural networks. The extracted Boolean functions make the hidden units understandable. However, in some cases, the extracted Boolean functions are complicated, and so are not understandable, which means that the hidden units are not functionally localized. This paper presents an algorithm for the functional localization of (the hidden units of) neural networks. When a hidden unit is well approximated to a low-order Boolean function, the unit can be regarded as functionally localized. The functional localization of a hidden unit is evaluated by the error between the hidden unit and the low-order Boolean function extracted from the hidden unit. The optimization is executed by genetic algorithms. We applied it to vote data, mushroom data and chess data. Experimental results show that the algorithm works well.

Rule Discovery from fMRI Brain Images by Logical Regression Analysis

This paper presents rule discovery from fMRI brain images. The algorithm for the discovery is the Logical Regression Analysis, which consists of two steps. The first step is regression analysis. The second step is rule extraction from the regression formula obtained by the regression analysis. In this paper, we use nonparametric regression analysis as a regression analysis, since there are not sufficient data in rule discovery from fMRI brain images. The algorithm was applied to several experimental tasks such as finger tapping and calculation. This paper reports the experiment of calculation, which has rediscovered well-known facts and discovered new facts.

Pattern Reasoning: A New Solution for Knowledge Acquisition Problem

This paper presents pattern reasoning, that is, the logical reasoning of patterns. Pattern reasoning is a new solution for the knowledge acquisition problem. Knowledge acquisition tried to acquire linguistic rules from patterns. In contrast, we try to modify logics to reason patterns. Patterns are represented as functions, which are approximated by neural networks. Therefore, pattern reasoning is realized by logical reasoning of neural networks. A few non-classical logics can reason neural networks, because neural networks can be basically regarded as multilinear functions and the logics are complete for multilinear function space, therefore, the logics can reason neural networks. This paper explains intermediate logic LC as an example of the logics and demonstrates how neural networks can be reasoned by LC.