December 27th, 2012, 06:47 AM
Recognizing Hand-writing Numbers
How can we teach computers to recognize hand-writing numbers with artificial neural networks? Any useful ideas, methods or tutorials?
December 27th, 2012, 07:12 AM
Originally Posted by freakyboard
There is a tutorial for Neural Networks (backpropagation algorithm) on the Internet. Since I am knew here, I cannot provide any URL (part of rules).
It talk about the EBP (backpropagation algorithm) and etc and even provided an application in Java (no source code provided).
Perhaps you already know enough about this subject and this site won't help you and just want to know how to teach computers to recognize handwriting numbers using neural networks.
December 28th, 2012, 03:10 AM
I remember choosing the parameters for your network is difficult. Eg the number of hidden layers and number of neurons per layer. This has an impact on the 'generalisation'. When you use lots and lots of neurons the network will be too flexible. (comparable to a model that follows the noise on the data)
Suppose you see some digits (the training data), if your parameters are off, the chance exists that you will only recognise the digits you saw and not recognise different digits. This is called overfitting and it's a pain :)