Tabular and Deep Learning of Whittle Index

Francisco Robledo (1, 2) , Urtzi Ayesta , Konstantin Avrachenkov (3) , Vivek S Borkar (4)

1 UPV / EHU - University of the Basque Country = Euskal Herriko Unibertsitatea
2 LMAP - Laboratoire de Mathématiques et de leurs Applications [Pau]
3 NEO - Network Engineering and Operations
4 EE-IIT - Department of Electrical Engineering [IIT-Bombay]

- Whittle index policy is an asymptotically optimal heuristic for solving Restless Multi-Armed Bandit Problems (RMBAP). - We propose two algorithms, QWI and QWINN, for the computation of such indices. - Both employ a two timescale system for the computation of the indices and the Q-values of each state/action.

Domaines: Informatique [cs]; Apprentissage [cs.LG]; Intelligence artificielle [cs.AI]

Francisco Robledo, Urtzi Ayesta, Konstantin Avrachenkov, Vivek S Borkar. Tabular and Deep Learning of Whittle Index. EWRL 2022 - 15th European Workshop on Reinforcement Learning, Sep 2022, Milan, Italy. ⟨hal-03810695⟩