Week of 11.14

Week of 11/14

     This week, we went to CalTech and learned about recommender systems. Ethan the graduate student explained how recommender systems operated in services like Netflix, Spotify, YouTube, etc. I thought it was pretty interesting how these systems worked, however, I'm still a bit perplexed by how it works and what we're supposed to do with the knowledge we have about it. There's a lot of math involved, and I'm not sure if I completely understand it. The closest thing I have got to understanding the system, however, is the Andrew Ng video series:

     To be completely honest with y'all out here, I feel really lost for the concept map. I don't really understand what's going on and these complicated math stuffs really feels like it's just a whole different language. Everyone else seems to understand it, but I think I may need to take some more cracks at it before I can decipher the whole variable soup. Ethan tried to break it down into simpler terms for us, and the things he said made sense to me. But beyond that, I'm really legitimately confused. 

     Sweet christ, I just realized that it doesn't look like there's a lot of writing on this. But we did spend a lot of time brainstorming our plans for our other survey project we've been thinking of, and how we plan to anonymize it. The board basically says that the recommender system we can use right now can capitalize on feature extraction. We feed the computer a matrix of users and their preferences in regards to certain movies. Using some mysterious algorithms (still figuring that out), the computer can give us two matrices that multiply to that matrix we fed it. These two matrices are pretty important.

   One of the matrices plots the relation between users and their preferences for some unknown features. The number of features are minimized so that there's not too many repeats in data (i.e. people liking Robert Downey Jr. and liking superheroes), and values are assigned as 0 and 1's. You actually don't know what the specific features are, but all you would know would be that there are features and similarities that exist. The second matrix is similar in the sense that it matches features to movies. It is in this arrangement that the two matrices can multiply together to form that one matrix you've already fed into the computer.

     There obviously still is a lot of things I don't really understand about this whole she-bang, but I think we tried to figure it out on our own? Yeah, no, we're still confused.

     On another note, I'm supposed to be working on the shirt designs. I have no idea what I'm doing, but I'll just mix-and-match some of the designs:

     I'm not sure which ones to choose just yet, but I might have an idea of what I want to do. We'll see...

    By the way, here's my video from the last concept map, I hope it's decent: