Online coordinate boosting

We present a new online boosting algorithm for updating the weights of a boosted classifier, which yields a closer approximation to the edges found by Freund and Schapire´s AdaBoost algorithm than previous online boosting algorithms. We contribute a new way of deriving the online algorithm that ties together previous online boosting work. The online algorithm is derived by minimizing AdaBoost´s loss as a single example is added to the training set. The equations show that the optimization is computationally expensive. However, a fast online approximation is possible. We compare approximation error to edges found by batch AdaBoost on synthetic datasets and generalization error on face datasets and the MNIST dataset.