Diagonalization of time-delayed covariance matrices does not guarantee statistical independence in high-dimensional feature space
نویسندگان
چکیده
Independent Slow Feature Analysis (ISFA) is an algorithm for performing nonlinear blind source separation, which combines linear ICA with Slow Feature Analysis (SFA). In its current form the objective function is based on time-delayed covariance matrices. While the algorithm performs well in general, we occasionally encountered cases in which the estimated sources are highly statistically dependent. Here we present a detailed analysis of these cases, which has revealed that second-order covariance matrices do not guarantee statistical independence of a few signals extracted from a high-dimensional feature space.
منابع مشابه
Approximate Joint Diagonalization and Geometric Mean of Symmetric Positive Definite Matrices
We explore the connection between two problems that have arisen independently in the signal processing and related fields: the estimation of the geometric mean of a set of symmetric positive definite (SPD) matrices and their approximate joint diagonalization (AJD). Today there is a considerable interest in estimating the geometric mean of a SPD matrix set in the manifold of SPD matrices endowed...
متن کاملAn optimal kernel feature extractor and its application to EEG signal classification
An optimal nonlinear feature extractor for extracting energy features under two different kinds of patterns is proposed. It carries out the simultaneous diagonalization of two signal covariance matrices in a high-dimensional kernel transformed space, and thus promises to find features which are more discriminant, especially when the original data have nonlinear structures. Two operations, white...
متن کاملReply to Jurdy & Stefanick comment
We disagree with virtually all of what Jurdy & Stefanick have written. Part of our disagreement stems from personal opinions about what is ‘simple’, ‘arbitrary’, ‘artificial’, ‘undesirable’, etc., but other disagreements are more profound and reveal a very different understanding of finite rotations. Jurdy & Stefanick raise two basic objections. One concerns statistical questions that were not ...
متن کاملSmall Sample Problem in Bayes Plug-in Classifier for Image Recognition
Image pattern recognition problems, especially face and facial expression ones, are commonly related to “small sample size” problems. In such applications there are a large number of features available but the number of training samples for each pattern is considerably less than the dimension of the feature space. The Bayes plug-in classifier has been successfully applied to discriminate high d...
متن کاملFeature Selection Using Multi Objective Genetic Algorithm with Support Vector Machine
Different approaches have been proposed for feature selection to obtain suitable features subset among all features. These methods search feature space for feature subsets which satisfies some criteria or optimizes several objective functions. The objective functions are divided into two main groups: filter and wrapper methods. In filter methods, features subsets are selected due to some measu...
متن کامل