Source Separation and Density Estimation by Faithful Equivariant SOM

Part of Advances in Neural Information Processing Systems 9 (NIPS 1996)

Bibtex Metadata Paper


Juan Lin, Jack Cowan, David Grier


We couple the tasks of source separation and density estimation by extracting the local geometrical structure of distributions ob(cid:173) tained from mixtures of statistically independent sources. Our modifications of the self-organizing map (SOM) algorithm results in purely digital learning rules which perform non-parametric his(cid:173) togram density estimation. The non-parametric nature of the sep(cid:173) aration allows for source separation of non-linear mixtures. An anisotropic coupling is introduced into our SOM with the role of aligning the network locally with the independent component con(cid:173) tours. This approach provides an exact verification condition for source separation with no prior on the source distributions.