Complexity Pursuit for Unifying Model

摘要

Complexity pursuit is an extension of projection pursuit to time series data and the method is closely related to blind separation of time-dependent source signals and independent component analysis. The goal is to find projections of time series that have interesting structure, defined using criteria related to Kolmogoroff complexity or coding length. In this paper, we first derive a simple approximation of coding length for unifying model that takes into account nongaussianity of sources, their autocorrelations and their smoothly changing nonstationary variances. Next, a fixed-point algorithm is proposed by using approximate Newton method. Finally, simulations verify the fixed-point algorithm converges faster than the existing gradient algorithm and it is more simple to implement due to it does not need any learning rate.