/[gzz]/gzz/Documentation/Manuscripts/Paper/paper.tex
ViewVC logotype

Diff of /gzz/Documentation/Manuscripts/Paper/paper.tex

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.39 by tjl, Tue Nov 19 20:19:39 2002 UTC revision 1.40 by tjl, Tue Nov 19 20:35:10 2002 UTC
# Line 205  Features independent at XXX should not c Line 205  Features independent at XXX should not c
205  if all circles were green and all squares yellow, a considerable amount of  if all circles were green and all squares yellow, a considerable amount of
206  bits would be wasted.  bits would be wasted.
207    
208    To understand why it is possible to learn to discriminate particular
209    textures easily, consider the task of learning {\em one} texture.
210    This is a two-class problem.
211    Extensive literature...
212    Here,
213    there are two categories: the paper A and everything else.
214    This problem is quite easy to solve: the probability of error
215    decreases exponentially in the number of features.
216    
217    
218    A perceptron\cite{XXX} is a simple neural network with XXX.
219    In the input layer, various features are activated,
220    ... linear combination of features.
221    
222    Easy to learn when many features, STRICT CORRELATION
223    
224    $m$ binary features, $N$ vectors. Probability that
225    
226    
227  Easiest to remember presence and absence of features; therefore, should have  Easiest to remember presence and absence of features; therefore, should have
228  relatively small basis size, not to have too many features. XXX why?  relatively small basis size, not to have too many features. XXX why?
229    
# Line 685  N textures, product texture environment, Line 704  N textures, product texture environment,
704  4 textures, texture shading. These correspond to G400,  GeForce2 and  4 textures, texture shading. These correspond to G400,  GeForce2 and
705  GeForce3.  GeForce3.
706    
 \section{Experiment}  
   
 \section{A neurocomputing interpretation}  
   
 The recognizability of the generated textures is perhaps surprising  
 in the light of the experiments on XXX..  
   
 To understand why it is possible to learn to discriminate particular  
 textures easily, consider the task of learning {\em one} texture.  
 This is a two-class problem.  
 Extensive literature...  
 Here,  
 there are two categories: the paper A and everything else.  
 This problem is quite easy to solve: the probability of error  
 decreases exponentially in the number of features.  
   
   
 A perceptron\cite{XXX} is a simple neural network with XXX.  
 In the input layer, various features are activated,  
 ... linear combination of features.  
   
 Easy to learn when many features, STRICT CORRELATION  
   
 $m$ binary features, $N$ vectors. Probability that  
707    
708    
709  \section{Conclusions}  \section{Conclusions}

Legend:
Removed from v.1.39  
changed lines
  Added in v.1.40

savannah-hackers-public@gnu.org
ViewVC Help
Powered by ViewVC 1.1.26