Artificial Neural Nets

x011 at Lehigh.EDU x011 at Lehigh.EDU
Wed Apr 26 08:22:15 EST 1995


In article <3njd56$db0 at mozo.cc.purdue.edu>, gfrancis at wizard (Greg Francis) write
s:
>> In article <3ngtbp$pgs at news.tuwien.ac.at>, e8627164 at fbma.tuwien.ac.at (Otto H
ain
>> zl) writes:
>> >Daniel Rabinovitch (drabinov at cs.cornell.edu) wrote:
>> >I think the flow is from natural NN's to artificial NN's but never back.
>> >This means that the artificial neuronal networks are modeled after there
>> >natural counterpart but not for giving feedback on the research done there.
>> >Same with genetic algorithms.
>> >
>> >Otto
>
>This is an interesting claim. A colleague and I were discussing just last week
that
>we could not think of a single example (outside of visual perception) where stu
dies
>of real neuronal networks have lead to a better understanding of a specific asp
ect
>of human cognitive behavior.
>
>Can anybody give me examples where neurophysiology studies produced better (or
even
>different) theories of cognition? (I will not accept general claims of modulari
ty,
>distributed coding, and the like, I am looking for something specific.)
>
>It seems to me that the direction is mostly the _reverse_ of the above claim.
>Researchers look for neurophysiological evidence of cognitive theories (whether
 they
>be guised as neural networks or otherwise).
>--
>Greg Francis, PhD    | "It's the opposite of fun. It's golf."
>Cognitive Psychology | Ellen Degeneres, "These Friends Of Mine"
>Purdue University    | http://www.psych.purdue.edu/cognitive.html
>Assistant Professor  | NeXTMail OK.
>
Jonathan A. Marshall and Richard K. Alley reported in their
article "A Self-Organizing Neural Network that Learns to Detect and
Represent Visual Depth from Occlusion Events"
and  "Adaptive Perceptual Pattern Recognition by Self-Organizing Neural
Networks: Context, Uncertainty, Multiplicity, and Scale"  archived at
ftp archive.cis.ohio-state.edu
    cd pub/neuroprose

that they had used a neuro net program to learn depth perception with
and on channel and off channel.  The learning rule uses disinhibitory
signals emitted from the on channel to trigger learning in the off
channel.

The neural network "EXIN" self organize in excitatory + inhibitory
networks.  The new learning rule allows for superposition of
multiple simultaneous neural activation.

The above program has been used to train arm movement and hearing through
echo location.



More information about the Neur-sci mailing list