Boosting Multiclass Learning With Repeating Codes

A long-standing goal of machine learning is to build a system which can detect a large number of classes with accuracy and efficiency. Some relationships between classes would become a scale-free network in which the authors can classify the assigned class very fast. Many available methods for multiclass problems have been proposed in the literatures, such as AdaBoost.ECC, AdaBoost.ERP and JointBoost. However, many of them are inaccurate or time-consuming on training. In this paper, they propose a new algorithm, called AdaBoost.ERC, which combines the approach of Dietterich and Bakiri based on Error Correcting Output Codes (ECOC) and Shapire's boosting algorithm.