DHARMA: Discriminant hyperplane abstracting residuals minimization algorithm for separating clusters with fuzzy boundaries

The problem of learning the discriminant hyperplanes, given imperfectly supervised training sample sets (which include unreliably labeled samples along the joint boundaries between the sample clusters), represents the topic of this study. The approach is to view the problem as the classical linear inequality problem, but subject to certain additional minimization constraints, and convert it into an equivalent unconstrained linear inequality problem, which is then solved through one of the established procedures in this field.