Self-Paced Co-training

Notation and Definition: We assume that examples are drawn from some distributions D over an instance space X = X1 × X2, where X1 and X2 correspond to two different “views” of examples. Let c denote the target function, and let X and X− (for simplicity we assume we are doing binary classification) denote the positive and negative regions of X , respectively . For i ∈ 1, 2, let X i = {xj ∈ Xi : ci(xj) = 1}, so we can think of X as X 1 ×X + 2 , and let X − i = Xi −X + i . Let D + and D− denote the marginal distribution of D over X and X−, respectively.