Learning From Weakly Supervised Data by The Expectation Loss SVM (e-SVM) algorithm

In many situations we have some measurement of confidence on "positiveness" for a binary label. The "positiveness" is a continuous value whose range is a bounded interval.