Delaytron: Efficient Learning of Multiclass Classifiers with Delayed
Bandit Feedbacks

In this paper, we present online algorithm called {\it Delaytron} for learning multi class classifiers using delayed bandit feedbacks. The sequence of feedback delays is unknown to the algorithm. At the -th round, the algorithm observes an example and predicts a label and receives the bandit feedback only rounds later. When , we consider that the feedback for the -th round is missing. We show that the proposed algorithm achieves regret of when the loss for each missing sample is upper bounded by . In the case when the loss for missing samples is not upper bounded, the regret achieved by Delaytron is where is the set of missing samples in rounds. These bounds were achieved with a constant step size which requires the knowledge of and . For the case when and are unknown, we use a doubling trick for online learning and proposed Adaptive Delaytron. We show that Adaptive Delaytron achieves a regret bound of . We show the effectiveness of our approach by experimenting on various datasets and comparing with state-of-the-art approaches.
View on arXiv