大阪大学　数理・データ科学教育研究センター
Center for Mathematical Modeling and Data Science,The University of Osaka

Multi-armed bandit problems under periodic observations

山崎和俊（University of Queensland）

1. ホーム
2. セミナー・集中講義
3. セミナー
4. Multi-armed bandit problems under periodic observations

大阪大学 数理・データ科学セミナー 金融・保険セミナーシリーズ 第154回

Multi-armed bandit problems under periodic observations

山崎和俊（University of Queensland）

We consider a version of the continuous-time multi-armed bandit problem in which decision opportunities occur at renewal times, and study its Gittins index policy. When the rewards are driven by Lévy processes and decision times follow a Poisson process, the Gittins index can be expressed explicitly in terms of a version of the Wiener–Hopf factorization, and it is shown to converge to that in the classical Lévy bandit model of Kaspi and Mandelbaum (1995). The diffusion and reflected Lévy cases are also examined.