Exécution de notre outil de synthèse sur un nouvel article
Vérifiez l’article ci-dessous, puis lancez sa synthèse.
Reinforcement Learning: Model Based
Shie Mannor (Technion – Israel Institute of Technology), Yishay Mansour (Tel Aviv University), Aviv Tamar (Technion – Israel Institute of Technology)
Résumé : This chapter develops model-based RL approaches where agents explicitly learn MDP transition and reward models from experience. The effective horizon concept quantifies how many steps matter significantly. Off-policy learning with generative models is analyzed first. On-policy learning requires explicit exploration strategies. Several algorithms are presented, including explicit explore-or-exploit, R-MAX and the PAC-MDP framework that ensures probably approximately correct (PAC) policies.