Multiclass Boosting: Simple and Intuitive Weak Learning Criteria

We study a generalization of boosting to the multiclass setting. We introduce a weak learning condition for multiclass classification that captures the original notion of weak learnability as being "slightly better than random guessing".