Massively Parallel Classification of EEG Signals Using Min-Max Modular Neural Networks

This paper presents a massively parallel method for classifying electroencephalogram (EEG) signals based on min-max modular neural networks. The method has several attractive features. a) A largescale, complex EEG classification problem can be easily broken down into a number of independent subproblems as small as the user needs. b) All of the subproblems can be easily learned by individual smaller network modules in parallel. c) The classification system acts quickly and facilitates hardware implementation. To demonstrate the effectiveness of the proposed method, we perform simulations on a set of 2,127 non-averaged single-trial hippocampal EEG data. Compared with a traditional approach based on multilayer perceptrons, our method converges very much faster and recognizes with high accuracy.