Federated Collaborative Networks

Jafar Tanha presented his paper “An AdaBoost Algorithm for Multiclass Semi-Supervised Learning”on the 2012 IEEE 12th International Conference on Data Mining in Brussels, Belgium. Please find the abstract below and the complete paper here.

We present an algorithm for multiclass Semi-Supervised learning which is learning from a limited amount of labeled data and plenty of unlabeled data. Existing semisupervised algorithms use approaches such as one-versus-all to convert the multiclass problem to several binary classification problems which is not optimal. We propose a multiclass semisupervised boosting algorithm that solves multiclass classification problems directly. The algorithm is based on a novel multiclass loss function consisting of the margin cost on labeled data and two regularization terms on labeled and unlabeled data. Experimental results on a number of UCI datasets show that the proposed algorithm performs better than the stateof-the-art boosting algorithms for multiclass semi-supervised learning.