So for CMPT 419, the artifical intelligence special topics class with machine learning as the subject, we have chanced from doing a final exam to a final project, something that usually is a “yay!” thing for me.

However, I am absolutely stumped on ideas of what I can do. I will most likely be leaning towards an application-type project.

I was thinking of making use of this as an opportunity to work on stuff for my little Web-based game project, but during brainstorming, I realized that I could not possibly do that because I would need to create the environment for the game to run first, and then actually use said environments to make something useful, so that is out of the question.

Some standard things, like getting a set of data and doing some classification or regression on that data has already been done in previous assignments for the class. Frankly, if I was going to do the project just like that, it would probably have been not enough work anyway.

A really crazy idea that is currently the front runner, simply because nothing else really stands out, is a machine learning project that takes pictures and applies taglines/captions for them. Not just any picture mind you, but pictures of this nature:

That’s right. This project will dabble into the “dark arts” of Internet memes and lolcats and lolruses and other crazy things to come up with things to label pictures as, such as “HALP” (pictured above), “im in ur X, Ying your Z“, “i has a X“, and so on. Silly idea and hated by some, but should I end up being extremely desperate, I would probably end up doing this.

Another one could be doing something with external ballistics. However, this is kind of tricky since (a) I can’t really say I’m brilliant with physics and (b) I don’t know much about ballistics in the first place. There also is no practical way for me to get data for such a thing, seeing how I don’t own firearms, airsoft guns, or paintball markers. It’s just interesting when you see projects like this really amazing paintball sentry.

Something else could be doing some machine learning to determine the value of a home in an area, given common evaluators like floor space, number of stories, age, building type, real estate type, number of bedrooms, number of bathrooms, and so on. However, in order to get the data I would need for such a project, I would need to do some data scraping which, according to the site’s terms of use, is illegal. If I end up doing this one, I would probably need to send those folks an email requesting permission to obtain the useful information I would need to create something. Or perhaps do something slightly reversed on that: given the price, number of bathrooms/bedrooms, and floor space, determine where this home is located (a classification problem).

Just some ideas that have been floating around and keeping me up at night. But if I am to get this thing done, I need to decide very, very soon just what I will be doing, as I do not have much time left.

This entry was posted on Thursday, July 31st, 2008 at 11:18 pm and is filed under School. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.