Neural Networks Project? [closed]

As it currently stands, this question is not a good fit for our Q&A format. We expect answers to be supported by facts, references, or expertise, but this question will likely solicit debat开发者_运维知识库e, arguments, polling, or extended discussion. If you feel that this question can be improved and possibly reopened, visit the help center for guidance. Closed 9 years ago.

I'm looking for ideas for a Neural Networks project that I could complete in about a month or so. I'm doing it for the National Science Fair, so I need something that has some curb appeal as well since it's being judged.

It doesn't necessarily have to be completely new and unique, I'm just looking for ideas, but it should be complex enough that it would impress someone who knows about the field. My first idea was to implement a spam filter of sorts, but I recently found out that NN's aren't a very good way to do it. I've already got a basic NN simulator with Genetic Algorithms, and I'm also adding the the generic back-propagation algorithms as well.

Any ideas?

---

Look into Numenta's Hierarchical Temporal Memory (HTM) concept. This may be slightly off topic if the expectation is of "traditional" Neural Nets, but it is also an extremely promising avenue for Artificial Intelligence.

Although Numenta introduced HTM and its associated software platform, NuPIC, almost five years ago, the first commercial product based upon this technology was released (in beta) a few weeks ago by Vitamin D. It is called Vitamin D Video and essentially turns any webcam or IP camera into a sophisticated video monitoring system, recognizing classes of items (say persons vs. cats or other animals) in the video feed.

With the proper setup, this type of application could make for an interesting display at the Science Fair, one with much "curb appeal".

To wet your appetite or even get your feet wet with HTM technology you can download NuPIC and check its various sample applications. Chances are that you may find something that meets typical criteria of both geekness and coolness for science fairs.

Generally, HTMs aim at solving problems which are simple for humans but difficult for computers; such a statement is somewhat of a generic/applicable to Neural Nets, but HTMs take this to the "next level".

Although written in C (I think) NuPIC is typically interfaced in Python, which makes it a convenient test bed for simple yet sophisticated proofs of concept applications.

---

You could always try to play around with a neural network and stock courses, if I had a month of spare time for a neural network implementation, thats what I would play with.

---

A friend of mine in college wrote a NN to play go on a 9x9 board.

I don't think it ever got very good, but I think it would be fun to try.

---

Look on how a bidirectional associative memory compare with other classical edit distance algorithms (Levenshtein, Damerau-Levenshtein etc) for typo correction. Also consider the articles on hebbian unlearning while training your NN - it seems that the confabulation phenomena is avoided.

---

I've done some works on top of NN, mainly an XML based language (Neural XML). See details here

http://amazedsaint.blogspot.com/search/label/Neural%20Network

Also, one interesting .NET Neural network project is Aforge.net - Check out that as well..

---

You can implement the game Cellz or create a controller for it. It was first created by Simon M Lucas. It's a nice and interesting game, and i'm sure that everyone will love it. I used it also for a school project and it turned out very ok.

You can find in that page some links to other interesting games.

---

How about applying it to predicting exchange rate (USD - EUR for example for sub minute trading) should be fun to show net gain of money over 1 month. I doubt this will work for trades longer than a minute... without a lot of extra work.

I like using committee machines so why not apply it to Face-Detection in images / movies or voice print authentication.

Finally you could get it to play pleasing music and use a crowd sourcing fitness function whereby people vote for the best "musicians"