强化学习浅介-实验室.pdf

强化学习浅介 俞扬 南京⼤学 计算机系
 软件新技术国家重点实验室
 机器学习与数据挖掘研究所 (LAMDA) Reinforcement learning intelligent animals can learn from interactions to adapt to the environment can computers do similarly? reinforcement learning, the “real artificial intelligence” action/decision Agent Environment reward state VALSE 2017 ：强化学习浅介 . Reinforcement learning action/decision Agent Environment reward state Agent’s inside: Policy: ⇡ : S ⇥ A ! R, Xa2A ⇡ (a|s) = 1 Policy (deterministic): ⇡ : S ! A Agent’s goal: learn a policy to maximize long-term total reward X X T 1 t T-step: t=1 rt discounted: t=1 rt VALSE 2017 ：强化学习浅介 . Difference between RL and SL? both learn a model ... supervised learning reinforcement learning environment environment data algorithm data algorithm