Bandits Dueling on Partially Ordered Sets

Julien Audiffren , Liva Ralaivola (1, 2)

1 QARMA - éQuipe d'AppRentissage de MArseille
2 IUF - Institut universitaire de France

We address the problem of dueling bandits defined on partially ordered sets, or posets. In this setting, arms may not be comparable, and there may be several (incomparable) optimal arms. We propose an algorithm, UnchainedBandits, that efficiently finds the set of optimal arms —the Pareto front— of any poset even when pairs of comparable arms cannot be a priori distinguished from pairs of incomparable arms, with a set of minimal assumptions. This means that Un-chainedBandits does not require information about comparability and can be used with limited knowledge of the poset. To achieve this, the algorithm relies on the concept of decoys, which stems from social psychology. We also provide theoretical guarantees on both the regret incurred and the number of comparison required by UnchainedBandits, and we report compelling empirical results.

Domaines: Apprentissage [cs.LG]; Machine Learning [stat.ML]

Julien Audiffren, Liva Ralaivola. Bandits Dueling on Partially Ordered Sets. Neural Information Processing Systems, NIPS 2017, Dec 2017, Long Beach, CA, United States. ⟨hal-01774844⟩