Artificial Intelligence Project

Question

Hi, all

I'm searching for an idea regarding AI term project. I've researched a lot about AI areas, but not able to decide/come up with reasonable project that can be implemented within 2 months (approx. 5-10 hours a week) and easy to test&document, but at the same time be non-trivial (to implement).

Do you guys have any ideas, previous projects, future project that meet this criteria?

Note, I'm not interested in ready solutions/implementation, but in ideas or project specification.

Thanks a lot. Regards.

Answer 1

Have you tried to look for papers related to AI? Because it is a general area. My suggestion is that you have to look on google scholar, ACM, etc. for papers related to AI and you can find some project ideas. In addition, you can give more keywords when you will be searching. For example, Handwritten recognition(more related to Machine Learning) or projects related to NLP (Information Retrieval, Textual Entailtment, Named Entity Recognition, etc.)

Answer 2

It's somewhat meta, but here's a thought: identifying a good heuristic for what qualifies as a "good" question and answer coupling on Stack Overflow. While this might seem obvious - just consider the question and answer up votes - I would submit to you consider the following data in an algorithm:

1. How common are the question's tags relative to the universe of questions and tags? So couplings whose "up" votes are high relative to the commonality of the topic should be considered better. Then again how do you factor in down votes? Arguably, a question with 10 up votes and no down is better than one with 15 up and 5 down.
2. Take the reputation of the answerers into account. Perhaps also use what their reputation is relative to the tags.
3. Look at how the question holds up over time. What votes are awarded say 72 hours after the fact, 30 days, or 90 days for instance? Maybe consider weighting those more highly because lurkers are no longer jumping on it.

I think you could imply other ideas from the tag system already in place on Stack Overflow, but I think it's possible to develop a more nuanced and potentially better system. I would describe it as the Netflix challenge for Stack Overflow in a sense.

Disclaimer: I don't have any relationship whatsoever with anyone who runs this site. Nor am I in any position to benefit from the outcome of this idea more so than any other user.

Answer 3

Why don't you try to solve a maze using autonomous robot. I use it in my Masters AI project. I used Lego mindstroms robot and RobotC comppiler to program it. http://mindstorms.lego.com/en-us/Default.aspx http://www.robotc.net/