Here are my real thoughts. Computer vision is a way for me to build; it's not the end all be all area that I'm deadset on, but it's certainly the best way that I can learn to apply my skills and build real stuff. Now that I have an actual job where I get to build for people behind the scenes, it's going to be hard to leave. I'm still reading up on literature (see readings), and I'm getting to the point where I can begin to apply my knowledge to push the frontier forward. Both inside of computer vision, and out.
Beyond the frontier, I'm looking to take advantage what I'm interested in (ie; baseball). The stuff on my website right now only scratches the surface, a small amalgamation of my thoughts. In short, however, I'm looking for ways to convert by heuristic biases into models in the hopes that one of them works. Various momentum and value strategies later, I'm happy to say that one of them works ;) I highly recommend reading the AQR paper below to learn about these ideas, because they're applicable to a lot more than baseball.