電子情報通信学会ソサイエティ大会講演要旨
BS-7-11
Vanishing Component Analysis for Nonlinear Blind Source Separation
◎LU WANG・TOMOAKI OHTSUKI(Keio Univ.)
Nonlinear blind source separation (BSS) is one of the biggest unsolved problems in unsupervised learning. In this report, we present a novel approach based on VCA for nonlinear BSS, which is referred to as VCA-based temporal decorrelation source separation (VCTDSEP). The algorithm relies on a new mathematical construction that permits to find a set of polynomials and to adapt to the form and number of polynomials, which guarantees the robustness to parameters. The simulation results show that it is able to recover a nonlinear mixture of two audio signals with a high reliability.