EcoICA: Skewness-based ICA via Eigenvectors of Cumulant Operator
نویسندگان
چکیده
Independent component analysis (ICA) is an important unsupervised learning method. Most popular ICAmethods use kurtosis as a metric of non-Gaussianity to maximize, such as FastICA and JADE. However, their assumption of kurtosic sources may not always be satisfied in practice. For weak-kurtosic but skewed sources, kurtosis-based methods could fail while skewness-based methods seem more promising, where skewness is another non-Gaussianity metric measuring the nonsymmetry of signals. Partly due to the common assumption of signal symmetry, skewness-based ICA has not been systematically studied in spite of some existing works. In this paper, we take a systematic approach to develop EcoICA, a new skewness-based ICA method for weak-kurtosic but skewed sources. Specifically, we design a new cumulant operator, define its eigenvalues and eigenvectors, reveal their connections with the ICA model to formulate the EcoICA problem, and use Jacobi method to solve it. Experiments on both synthetic and real data show the superior performance of EcoICA over existing kurtosis-based and skewness-based methods for skewed sources. In particular, EcoICA is less sensitive to sample size, noise, and outlier than other methods. Studies on face recognition further confirm the usefulness of EcoICA in classification.
منابع مشابه
Eigensubspace Algorithms for Estimating the Polyspectral Parameters of Harmonic Processes
e-mail : [email protected] The polyspectral parameters of a harmonic process are defined by the locations and strengths of the polyspectral impulses in the higher dimensional frequency space. MUSIC and ESPRIT-like algorithms for extracting these parameters, when the signal is corrupted by coloured Gaussian noise of unknown statistics, are proposed in this paper. The MUSIC -like algorith...
متن کاملEigendecomposition of self-tuned cumulant-matrices for blind source separation
Existing algorithms for blind source separation are often based on the eigendecomposition of fourth-order cumulant matrices. However, when the cumulant matrices have close eigenvalues, their eigenvectors are very sensitive to errors in the estimation of the matrices. In this paper, we show how to produce a cumulant matrix that has a well-separated extremal eigenvalue. The corresponding eigenvec...
متن کاملProper Inner Product with Mean Displacement for Gaussian Noise Invariant ICA
Independent Component Analysis (ICA) is a classical method for Blind Source Separation (BSS). In this paper, we are interested in ICA in the presence of noise, i.e., the noisy ICA problem. Pseudo-Euclidean Gradient Iteration (PEGI) is a recent cumulant-based method that defines a pseudo Euclidean inner product to replace a quasi-whitening step in Gaussian noise invariant ICA. However, PEGI has ...
متن کاملESTIMATING EQUATIONS FOR SOURCE SEPARATION Jean-Fran
This paper proposes a unifying view of source separation via the concepts of `estimating function' and `estimating equation'. We exhibit the estimating functions corresponding to various known techniques like ICA, JADE, infomax, maximum likelihood, cumulant matching, etc: : :We also show how equivariant batch and adaptive algorithms stem from each particular estimating function and discuss thei...
متن کاملBlind Channel Identification in (2 ×1) Alamouti Coded Systems Based on Maximizing the Eigenvalue Spread of Cumulant Matrices
Channel estimation in the (2×1) Alamouti space-time block coded systems can be performed blindly from the eigendecomposition (or diagonalization) of matrices composed of the receive antenna output 4th-order cumulants. In order to estimate the channel, we will propose to choose the cumulant matrix with maximum eigenvalue spread of cumulant-matrix. This matrix is determined in closed form. Simula...
متن کامل