On Multilabel Classification and Ranking with Partial Feedback
This work addresses the challenge of efficient learning in multilabel settings with limited feedback, which is incremental as it builds on prior methods to enhance regret bounds and practical performance.
The paper tackles the problem of multilabel classification and ranking with partial feedback by introducing a novel algorithm based on second-order descent methods and upper-confidence bounds for exploration-exploitation trade-offs, achieving O(T^{1/2} log T) regret bounds that improve on existing results and showing comparable performance to full-information baselines on real-world datasets.
We present a novel multilabel/ranking algorithm working in partial information settings. The algorithm is based on 2nd-order descent methods, and relies on upper-confidence bounds to trade-off exploration and exploitation. We analyze this algorithm in a partial adversarial setting, where covariates can be adversarial, but multilabel probabilities are ruled by (generalized) linear models. We show O(T^{1/2} log T) regret bounds, which improve in several ways on the existing results. We test the effectiveness of our upper-confidence scheme by contrasting against full-information baselines on real-world multilabel datasets, often obtaining comparable performance.