In this paper we consider a problem of source separation when sources are second-order non-stationary stochastic processes. We employ the natural gradient method and develop learning algorithms for both linear feedback and feedforward neural networks. Thus our algorithms possess equivariant property. Local stability analysis shows that separating solutions are always locally stable stationary points of the proposed algorithms, regardless of probability distributions of sources.
|Number of pages||5|
|Publication status||Published - 2002|
|Event||2002 International Joint Conference on Neural Networks (IJCNN '02) - Honolulu, HI, United States|
Duration: 12 May 2002 → 17 May 2002
|Conference||2002 International Joint Conference on Neural Networks (IJCNN '02)|
|Period||12/05/02 → 17/05/02|