... is to fail as much as necessary as soon as possible
Cutting Plane Algorithm for lp-Norm Multiple Kernel Learning: This code was written as part of the Convex Optimization Course Project along with Uma. Here, we implement the cutting plane algorithm from section 2.4.2 of this paper. [code]
Class Ratio Estimation: An implementation of my ICML 2014 paper is available in Java on github. The cutting plane algorithm for the kernel learning is incomplete since there is no SDP solver that I can directly use at the moment. However, kernel learning is solved using projected subgradient method in this implementation.