Wednesday, October 31, 2012

Welcome to my renewed blog!


My site was hacked two years ago and since then, I haven't bothered restoring the site.

Now, I'll start writing again. What's the theme this time? Still pretty technical but perhaps I'll place more emphasis on machine learning and related stuffs, which is what I'm doing now for a living. At the moment, I am personally working on few projects, including:
  1. A new efficient online logistic regression method, so-called dual regression. I've recently found a flaw in my approach which makes this project likely to fail. Anyway, I still believe that stochastic gradient descent, which is the current standard way to do online learning, can be significantly improved, both empirically and theoretically.
  2. Improve the Kernel SVM method by using a fast kernel summation method (also called fast multipole method or FMM in Computational Physics).
  3. Scale up and scale out certain ML algorithms.