a question: can neural networks be made to do this?

Rob Butera rbutera at is.rice.edu
Tue Jul 14 19:15:07 EST 1992


In article <1992Jul14.181707.3007 at athena.mit.edu>, mlevin at presto.cs.tufts.edu (Mike Levin) writes:
|> 
|>     I have a question about applying neural nets to a certain problem.|>
|>     ...
|> then giving it the correct seed to let it adjust its weights. The hope
|> then is that after a while the net's guesses will become good, and it
|> can be used to derive seeds for arbitrary images. There is an infinite
|> supply of training material, because there is an infinite amount of
|> randomly-chosedn seeds to produce images.
|>     So, does anyone have any suggestions about this? Is it feasable to
|> feed a network an image as input? What do I need to do to make it

Well, not necessarily using neural nets, but image compression reporduction
like this has been done.  The cover story in IEEE spectrum a couple of
years ago featured a couple of former math professors at Georgia Tech
(one was Barnesley, I THINK the other was Sloan).  They processed images so as
to reduce them to a few fractal-like seeds that through algorithmic 
iteration could be used to regenerate the image.  They didn't achieve
perfect reconstruction, but you could specify what parts of the image
had to be reproduced well and which didn't.  Anyway, check the back issues
for references. 

I have no clue if they used any kind of neural net.  I sure they didn't use
the "real" neural nets typically discussed on here  :)

-- 
Rob Butera - ECE grad student   Rice University, Houston, TX   rbutera at rice.edu

 "I do not feel obliged to believe that same God who endowed us with sense,
  reason, and intellect had intended for us to forgo their use."  - Galileo



More information about the Comp-bio mailing list