1. Lectures started this week! It's a bit odd only having nine hours of class time in a week, but there still seems to be plenty to do. I wonder how many of us are realistically putting in the other 3+ hours a week on project work.
2. Most of my project time this week has gone to watching video lectures on machine learning (via Coursera). One of the neatest techniques is an algorithm called gradient descent. For each iteration it considers each parameter A and a function J to minimise. Then it sets A to be A - c(dJ/dA) for a learning rate constant c. This means that at the top of a hollow, you move down the steep sides rapidly. As things shallow out towards the minimum, the steps get increasingly smaller. Awesome. (If you think of a derivative as the slope at a point, you can see this, as well as why the algorithm always moves towards the lowest point.)
3. Next week is going to be a toss up between machine learning videos and reading papers on things like Kalman filtering. Both are interesting and a little zombie-ish in that they tend to steal my brains.
4. I think students might be anti-zombies, since Statistical Physics, Quantum Mechanics and Partial Differential equations somehow have similar effects. Considering how many things are like that, it must be me, right? They can't all be zombie subjects.
5. Despite a bit of grumbling, this is totally what I signed up for and it's really great. Now I just need to get my brain into gear so I can stop confusing scalar products with scalar multiplication, which, despite the names, are not the same thing at all. Oops! I blame the zombies.
No comments:
Post a Comment