Heuristic search for EEG features by learning-type fuzzy template matching algorithm

Abstract

Thus, it is a problem that conventional BCI does not suit for subjects whose targetted EEG feature does not appear significantly. To solve the problem, we attempt development of heuristic BCI, which automatically extracts EEG patterns appearing at task run time. In this study, we proposed a method to search for characteristic EEG patterns during performing the task by learning, in which fuzzy template matching method that based on learning-base -simple-fuzzy-reasoning is employed in order to compose heuristic space limitedly in a fuzzy rule. As the first evaluation of this method, we evaluated the recognition of alpha-wave dominant state without promise information. We performed pruning that deletes rules with inadequate outputs by learning-failure in addition to the search by learning, and succeeded in the recognition of EEG states.