Support vector machines(SVMs) are state of the art classifiers, obtained by solving a convex Quadratic program defined over a training set of several (observation, label) pairs.
Motivated from real world problems, we consider the important issue of designing SVM like classifiers for handling uncertain observations.
In this talk we show that by using chance constraints one can derive tractable formulations which can yield classifiers robust to the uncertainty.
In particular the talk will discuss several SOCP formulations which can yield robust classification when only partial knowledge of uncertainty is available.
Back to Workshop IV: Robust Optimization