Bayesian Independent Component Analysis under Hierarchical Model on Independent Components | Toshiyasu Matsushima Laboratory

Title	Bayesian Independent Component Analysis under Hierarchical Model on Independent Components (in Japanese)
Authors	Kai Asaba 、Shota Saito 、Shunsuke Horii 、Toshiyasu Matsushima
Released Year	2018
Format	International Conference
Category	Knowledge information processing
Jounal Name	Proceedings of Asia-Pacific Signal and Information Processing Association Annual Summit and Conference
Jounal Page	pp.959--962
Published Year	2018
Published Month	11
Abstract （English）	Independent component analysis (ICA) deals with the problem of estimating unknown latent variables (independent components) from observed data. One of the previous studies of ICA assumes a Laplace distribution on independent components. However, this assumption makes it difficult to calculate the posterior distribution of independent components. On the other hand, in the problem of sparse linear regression, several studies have approximately calculated the posterior distribution of parameters by assuming a hierarchical model expressing a Laplace distribution. This paper considers ICA in which a hierarchical model expressing a Laplace distribution is assumed on independent components. For this hierarchical model, we propose a method of calculating the approximate posterior distribution of independent components by using a variational Bayes method. Through some experiments, we show the effectiveness of our proposed method.
Note （English）	1
Manuscript
Presentation