This is a simple site to share somethings I am interested, mainly programing related.
Any questions, feel free to contact me.
I have an idea that one day, most things can be optimized, and the most import and difficult thing is to find the right target to optimize, not how to optimize it.
Most problems in computer can be viewed as an optimization problem, esp. in data mining fields. Today, as far as I have learned, it is still not easy to optimize a general problem. We have a lot tools, like, LDA, PSI, SVM, Gaussian Process, Regression, Boosting. I find it's very fun to find respective optimization versions of them. Some of these are already presented as optimization problems, but some of them are not easy to do in optimization approaches.
There are two hard problems in optimization approach to most machine learning problems. First, there are no effective ways to find the global optimization point of a general problem. Second, there are no intuitive and direct ways to represent some domain experts knowledges as optimization function.
Thanks to Google Chrome NACL, I have an simple optimization implementation demo(This need to load a about 1MB native plugin), over which, we can try simple problems with optimization approaches.