关键词: 张量, 模式表示, 特征提取, 降维, 分类器设计

Abstract: Abstract: As an extension and complementarity of typical vector patterns, high order tensor patterns have recently attracted attention in the fields of machine learning and pattern recognition. The current tensorbased learning algorithms in terms of the following aspects are summarized: focusing on the existing problems of vector pattern representation, the necessity and importance of designing tensorized learning algorithms were discussed. From the facets of feature extraction and classifier design, some typical tensorbased learning algorithms were categorized. Moreover, their intrinsic connection, advantages and disadvantages were analyzed from the view of multilinear projections, essential differences and their own superiorities between tensor and vector patterns were clarified, and the problems to be solved and future research were discussed.

Key words: tensor, pattern representation, feature extraction, dimensionality reduction, classifier design