QUOTE(scalt @ 7 May, 2009 - 07:26 PM)

I have to do an assignment over the next month and a bit that involves coming with an idea for a good neural network project and then implementing that. I'm probably going to use python (and numpy) but MATLAB is also a possiblity.
Right now I'm kind of stuck on the idea for my project so if anyone out there has a few cool ideas about stuff you could build a neural network to solve/do then I'd love to hear them.
Previous years' projects included a paper/scissors/rock AI and a horse racing predictor.
Image-related projects are nice because the result is something one can actually see. For instance, I built a classifier which distinguishes "foliage" (leaves, grass, etc.) from "non-foliage", at the pixel level (
Pixel Classification Project - image links are down as of this writing). When the model is run over a new image, the result is a new image which is black where pixels are predicted to be "non-foliage" and white where pixels are predicted to be "foliage". Intermediate levels of response are indicated by shades of gray.
Skin detection is a popular topic of research, both commercially and academically. You could also try to discriminate among things which appear in common photographs like vacation snapshots, such as pavement, sky, etc.
-Will Dwinnell
Data Mining in MATLAB