Tony Bell, 4/25/03
<br>
<br>

Dependent Component Analysis - problems and some solutions
<br>
<br>

In order to build multi-layer self-organising
feature extraction networks, and solve the feedforward
part of the neural computation problem, we need to understand
the relationship between the nonlinear transformations
involved, the mathematics of the relevant feature spaces,
and the corresponding shapes in the probability 
distributions of the signals.

<br>
<br>

I will give an update on progress on this front, from the
point of view of an emerging class of algorithms which,
unlike ICA, search for simple forms of dependency structure
in natural data.