tech report available

[Tony Bell]

FTP-host: ftp.salk.edu 
FTP-file: pub/tony/bell.blind.ps.Z

The following technical report is ftp-able from the Salk Institute.
The file is called bell.blind.ps.Z, it is 0.3 Mbytes compressed,
0.9 Mbytes uncompressed, and 36 pages long (8 figures).

It describes work presented at NIPS '94, with various embellishments,
and a version of it will appear in Neural Computation in 1995.

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

           Technical Report no. INC-9501, February 1995, 
  Institute for Neural Computation, UCSD, San Diego, CA 92093-0523


             AN INFORMATION-MAXIMISATION APPROACH TO 
             BLIND SEPARATION AND BLIND DECONVOLUTION

              Anthony J. Bell & Terrence J. Sejnowski

               Computational Neurobiology Laboratory
                       The Salk Institute
                   10010 N. Torrey Pines Road
                   La Jolla, California 92037

                            ABSTRACT

    We derive a new learning algorithm which maximises the information 
transferred in a network of non-linear units. The algorithm does not 
assume any knowledge of the input distributions, and is defined here 
for the zero-noise limit. Under these conditions, information 
maximisation has extra properties not found in the linear case 
(Linsker 1989). The non-linearities in the transfer function are able 
to pick up higher-order moments of the input distributions and perform
true redundancy reduction between units in the output representation. 
This enables the network to separate statistically independent 
components in the inputs: a higher-order generalisation of Principal 
Components Analysis. 
    We apply the network to the source separation (or cocktail party) 
problem, successfully separating unknown mixtures of up to ten speakers. 
We also show that a variant on the network architecture is able to 
perform blind deconvolution (cancellation of unknown echoes and 
reverberation in a speech signal). Finally, we derive dependencies of 
information transfer on time delays. We suggest that information 
maximisation provides a unifying framework for problems in `blind' 
signal processing.

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

Can be obtained via ftp as follows:

  unix> ftp ftp.salk.edu  (or 198.202.70.34)

   (log in as "anonymous", e-mail address as password)

  ftp> binary
  ftp> cd pub/tony
  ftp> get bell.blind.ps.Z
  ftp> quit

  unix> uncompress bell.blind.ps.Z
  unix> lpr bell.blind.ps