Paper available: blind source separation

[Hagai Attias]

A new paper on blind separation of mixed and convolved sources 
is available at:

             http://keck.ucsf.edu/~hagai/papers.html


-------------------------------------------------------


             BLIND SOURCE SEPARATION AND DECONVOLUTION:
             THE DYNAMIC COMPONENT ANALYSIS ALGORITHM

              Hagai Attias and Christoph E. Schreiner
              University of California, San Francisco
                        hagai at phy.ucsf.edu

                (Neural Computation 1998, in press)


   We present a novel family of unsupervised learning algorithms for 
   blind separation of mixed and convolved sources. Our approach,
   termed `dynamic component analysis' (DCA), is based on formulating 
   the separation problem as a learning task of a spatio-temporal 
   generative model. The resulting learning rules achieve separation 
   by exploiting high-order spatio-temporal statistics of the observed 
   data. Using an extension of the relative-gradient concept to the      
   spatio-temporal case, we derive different rules by learning 
   generative models in the frequency and time domains, whereas a hybrid 
   frequency/time model leads to the best performance.

   These algorithms generalize independent component analysis (ICA)
   to the case of convolutive mixtures, and exhibit superior performance 
   on instantaneous mixtures. In Addition, our approach can incorporate
   information about the mixing situation when available, resulting in a 
   `semi-blind' separation algorithm. Finally, the spatio-temporal 
   redundancy reduction performed by DCA algorithms is shown to be 
   equivalent to information-rate maximization through a simple network.