状態空間モデルを用いた非線形独立成分分析

抄録

This paper proposes an independent component analysis method using state-space models. Firstly, a learning algorithm to estimate the parameters of the output function of the model is derived based on the property that the probability density function of the output of the model only depends on that of the input and the direct feedthrough term of the model. Secondly, the parameters of the dynamic equation are estimated by the information backpropagation method. Thirdly, since many systems such as mechanical systems do not have any direct feedthrough term, we extend the proposed algorithm to systems without direct feedthrough terms. Furthermore, a numerical simulation demonstrates the effectiveness of the proposed method.