Name: Firas Hamze <shibumi@stanford.edu>
Title: Neural network handwritten number recognition

	The project I intend to do is handwritten number recognition using
Neural Networks. When I was researching a project for another class, I ran
across a method developed by LeCun et al [1] for recognizing handwritten
zip-codes in the mail, and have been interested in trying something of the
sort since. It will be a way to learn about backpropagation feedforward
neural nets and their applications to image understanding.

	The project scope is quite straightforeward: using a set of
handwritten numbers (either my own one from a database) I will "train" a
set of neural network weights, and then observe the performance of the
weights on a set of test numerals. I will experiment with varying the
system parameters (such as number of neural net units, hidden layers,
system functions) and see how the classification accuracy changes.

	It was also possible to implement this system using the
eigenimages aproach, but I thought that I'd rather learn about this new
method. If time permits, I could compare the performace of the neural net
system to that of the eigenimages method, but since I'm working alone,
that may not be possible.

	The research and implementation should comprise the bulk of the
time of this project (probably more than half) with the report and
presentation preparation occupying the rest. Hopefully, I'll be already
under way by this weekend (I heard that neural nets can act peculiar
sometimes, so I'm taking extra time to be sure.)

	Reference:

[1] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W.
Hubbard, and L. D. Jackel. Backpropagation applied to handwritten zip code
recognition. Neural Computation, 1(4):541-551, Winter 1989.