Recently, new methods for measuring and analyzing ERP brain wave data are required since the continued growth in such large and complex data collection in both brain research and medical science. In order to discover various patterns hidden in ERP data, it is necessary to pay attention to two types of peculiarities: temporal (time) and spatial (channel), from the potential and gradient standpoints. In this paper, we propose a novel approach of POM (peculiarity oriented mining) based multi-aspect ERP brain wave data analysis. We describe how to design cognitive experiments on investigating human computation mechanism with multiple difficulty levels for obtaining multi-ERP data, and how to analyze and visualize spatiotemporal peculiarities of such data. The experimental results show that all objectives we expect for the approach are achievable.

View full abstract

Although Subsequence Time Series (STS) clustering has been one of the most popular techniques to extract typical subsequence patterns from time-series data, previous studies have gave surprising reports that cluster centers obtained using STS clustering closely resemble ``sine waves'' with little relation to input time-series data. This means that STS clustering cannot be used for its original purpose, extraction of typical subsequences. Despite this serious fact, its mathematical structure has seldom been studied. The main contribution of this paper is that we give a theoretical analysis of STS clustering from a frequency-analysis viewpoint and identify that sine waves are generated due to the superposition of time series subsequences, which have the same spectra but different phases. Another contribution is that we propose a clustering algorithm, which uses a phase alignment preprocessing, to avoid sine-wave patterns.

View full abstract

This paper proposes a method for the unsupervised learning of lexicons from pairs of a spoken utterance and an object as its meaning under the condition that any priori linguistic knowledge other than acoustic models of Japanese phonemes is not used. The main problems are the word segmentation of spoken utterances and the learning of the phoneme sequences of the words. To obtain a lexicon, a statistical model, which represents the joint probability of an utterance and an object, is learned based on the minimum description length (MDL) principle. The model consists of three parts: a word list in which each word is represented by a phoneme sequence, a word-bigram model, and a word-meaning model. Through alternate learning processes of these parts, acoustically, grammatically, and semantically appropriate units of phoneme sequences that cover all utterances are acquired as words. Experimental results show that our model can acquire phoneme sequences of object words with about 83.6% accuracy.

View full abstract