hmobahi at gmail.com wrote:
> Hi,
>> Has there ever been shown any correlation between hardness of learning
> a problem for human being and any artificial neural net model? For
> instance, we know that patterns whose features are not linearly
> separable are harder to be learned by an MLP. Has any one ever
> investigated this issue in human side to see if human has difficulty in
> classifying patterns whose boundary in not linearly separable? I am not
> talking about a general experiment; even a very specific experiment,
> say psychophysics, in a specified domain is also sufficient for my
> purpose.
Yes, there has been.
Start with these:
Predictability, complexity and learning. W Bialek, I Nemenman & N Tishby,
Neural Comp 13, 2409-2463 (2001).
http://www.princeton.edu/~wbialek/learning_links.html
Ilya Nemenman. Fluctuation-dissipation theorem and models of learning.
Neural Comp., 17(9):2006-2033, 2005.
http://www.menem.com/ilya/digital_library/full.html
for a solid theoretical background, then chase references therein for
experiments more specific to your questions.
-- Michael
