You would need a large database of chess games and a program that can detect some features in a position (for example: "white are a pawn up", "black pawn structure is better", "black king is not safe", etc.).
The idea would be to see if some features in a position are making one side more likely to win. For example, we might find that being a piece up makes one side more likely to win, since the largest majority of the cases when one side had a piece up, it ended up winning.
But that's not an interesting result. What I would like to see if it's possible to discover features that are good for one side, and which we didn't know about.
I didn't have time to start anything about it, though.
What I want is something different: mining tons of games to learn new stuff about positional strength in chess. For example, assume that we don't know that a queen is stronger than a knight, and instead think that the two pieces have the same value. We would trade those two pieces more or less indifferently, resulting in lots of games where one side has a queen and the other side has a knight. Now, the program I would like to do will mine all those games and say "99% of the cases where one side had a queen and the other side had a knight were a win for the side with the queen". And we would be able to learn that a queen is stronger to a knight... Now obviously this example is too simple, but I would like to see if it is possible to learn some more subtle things.
Just wondering out loud reg. your idea to a have a rick roll detector: How exactly would that work? A brute force approach would be to compare how similar the target video is to some sample rick roll videos, I am guessing music matching could be attempted since frame by frame duplicate detection of the video is insanely hard.
From a machine learning perspective, it gets harder because a good rick roll entices the target into clicking the link which turns out to be a false positive. On the other hand, the link could be completely normal so the features of the link itself or not necessarily useful. Hmmm...
http://cs229.stanford.edu/projects2012.html
Take any of these studies, and try to replicate or extend it. They're all pretty interesting and come with the advantage that you get to see how someone else approached the problem as well.
Speech recognition for cows,dogs,birds etc.
Moo,chirp,meow
and of course a siri plugin: hey siri, what sound/song is that?