A Novel Transductive Learning Algorithm Based on Multi-Agent-System
Jun Hui Pan · 2009
We consider the problem of multiclass classification where both a few labeled data and lots of unlabeled data are given, for which a new approach called Multi-Agent-System- Based Multi-Class Transductive Learning (MMT) is presented. In MMT, we transform the data classification into a self-organizing Markov stochastic process that finally converges to a stationary probability distribution, in which an optimal label distribution is provided. Based on the proposed approach, an algorithm called Multi-Agent-System-Based Multi-Class Transductive Algorithm (MMTA) was designed and its converging capabilities were discussed. The simulations have shown the effectiveness and practicability of MMTA.