台大-李宏毅-b站机器学习视频-课件rl.pptx

Deep Reinforcement LearningScratching the surfaceDeep Reinforcement LearningScenario of Reinforcement LearningActionObservationStateChange the environmentAgentDon’t do thatRewardEnvironmentScenario of Reinforcement LearningAgent learns to take actions to maximize expected reward. ActionObservationStateChange the environmentAgentThank you.RewardEnvironment/how-to-clean-site-structure/Learning to paly GoActionObservationNext MoveRewardEnvironmentLearning to paly GoAgent learns to take actions to maximize expected reward. ActionObservationRewardreward = 0 in most casesIf win, reward = 1If loss, reward = -1EnvironmentLearning to paly Go- Supervised v.s. ReinforcementLearning from teacherSupervised:Reinforcement LearningNext move:“5-5”Next move:“3-3”Learning from experienceFirst move…… many moves ……Win!(Two agents play with each other.)Alpha Go is supervised learning + reinforcement learning.Learning a chat-botSequence-to-sequence learningLearning a chat-bot- Supervised v.s. Reinforcement“Hello”Say “Hi”“Bye bye”Say “Good bye”………….…….Hello ?Bad……AgentAgentSupervised ReinforcementLearning a chat-bot- Reinforcement LearningHow old are you?How old are you?I am 16.See you.I though you were 12.See you.What make you think so?See you.Let two agents talk to each other (sometimes generate good dialogue, sometimes bad)Learning a chat-bot- Reinforcement LearningMachine learns from the evaluationDialogue 1Dialogue 2Dialogue 3Dialogue 4Dialogue 5Dialogue 6Dialogue 7Dialogue 8Deep Reinforcement Learning for Dialogue Generation /pdf/1606.01541v3.pdfBy this approach, we can generate a lot of dialogues.Use some pre-defined rules to evaluate the goodness of a dialogueMore applicationsUS President[Wu & Lee, INTERSPEECH 16]More precisely, please.TrumpIs it related to “Election”?userYes.I see!Here are what you are looking for.Interactive retrievalMore applicationsFlying Helicopter/watch?v=0JL04JJjoccDriving/watch?v=0xo1Ldx3L5QGoogle Cuts Its Giant Electricity Bill With DeepMind-Powered AI/new