Click to See Complete Forum and Search --> : A simple neural network, convergent problem?


choconlangthang
November 23rd, 2005, 04:03 AM
Hello there, plz help me with this.
I have a 3-layer feed-forward backpropagation neural network. I use it to recognize letter (printed uppcase only). The features extracted are 256 subregions of a 64x64pixel image (which means a region is 16x16 pixel). So the input vector contains 256 members.
However, when I try with 3 letters (A,B,C). It seems the product is one letter only, it is B. I am sure that my code is in correct implementation (the theory can be found here http://www.speech.sri.com/people/anand/771/html/node37.html ). I try to modify some parameters like learning rate, loop for each letter, but it seems there is no change.

Any idea is welcomed, plz help, thanks.

Pinky98
November 25th, 2005, 06:37 AM
Change your network size. If your network is too big or too small, it may have difficulty distinguishing between characters.