Abstract
We propose a novel dictionary-based learning method for ambiguously labeled multiclass classifification, where each training sample has multiple labels and only one of them is the correct label. The dictionary learning problem is solved using an iterative alternating algorithm. At each iteration of the algorithm, two alternating steps are performed: a confifidence update and a dictionary update. The confifidence of each sample is defifined as the probability distribution on its ambiguous labels. The dictionaries are updated using either soft (EM-based) or hard decision rules. Extensive evaluations on existing datasets demonstrate that the proposed method performs signifificantly better than state-of-the-art ambiguously labeled learning approaches.