Second Seminar Presentation
Download as PPTX, PDF0 likes519 views
The document discusses the use of case-based reasoning and reinforcement learning in developing adaptive intelligent systems for mobile experiences. It outlines the system architecture of an AI engine that incorporates various planning and execution modules, emphasizing the need for model-less learning and managing exploration-exploitation trade-offs. Additionally, it highlights a testing strategy based on the Turing test and references notable works in the field of AI and planning.
Second Seminar Presentation
- 2. AgendaRefreshment : Problems and GoalsAnswering the whyWhy we’ve used Case-Based Reasoning.Why we’ve used Reinforcement Learning.System Architecture.Project Testing StrategyTuring Test.NPC (Static AI).
- 6. Problems and GoalsMachines rely on static scripting techniques.AdaptiveAgentIntelligent
- 9. Problems and GoalsThe Absence of sharing experience costs a lot.MobileExperience
- 10. Case Based Reasoning- a Brief
- 12. Why Case-Based ReasoningPlan Learning
- 13. Why Case-Based ReasoningPlan LearningFailureLearning
- 14. Why Case-Based ReasoningPlan LearningFailureLearningCriticLearning
- 15. Why Case-Based ReasoningPlan LearningFailureLearningCriticLearningPrediction
- 16. Reinforcement Learning – A Brief
- 17. Why Reinforcement LearningRequires No ModelBalance Exploration- ExploitationApplies BootstrappingUsed in the Revising PhaseSub-optimal policies
- 18. Why Reinforcement LearningUsed in the Revising Phase
- 19. Why Reinforcement LearningRequires No Model
- 20. Why Reinforcement LearningApplies Bootstrapping
- 21. Why Reinforcement LearningLearn Sub-Optimal Policies
- 22. Why Reinforcement LearningBalance Exploration-Exploitation
- 23. System ArchitectureI-Strategizer AI Engine : Online Case Based PlannerI-StrategizerToWargusCase Based ReasonerEE ModuleGame StateGoalGame StateExpansion ModulePlan RetrieverPerception ModuleRetrieved PlanPlan to be adaptedAdapted Plan Game StatePlan AdaptorCase (Plan) BasePlan to be adaptedWargus (Game)Plan Game StateActions ExecutorGame Specific ActionsActionsExecuted PlanExecution ModulePlan Reviser(RL Techniques)Plan RetainerRevised PlanRetained PlanFeedbackGame Specific Feedback
- 24. Case Representation : An Example
- 25. Interleaved Expansion and Execution
- 26. Testing Strategy – Turing Test
- 27. Testing Strategy –Playing Static AI
- 28. ReferencesSantiago Ontanon, Ashwin Ram - On-Line Case based Planning– 2010KristianJ.Hammond - Case-Based Planning - A Framework for planning from Experience - 1994 Book: Reinforcement Learning An Introduction – 1998Matthew Molineaux, David W. Aha, & Philip Moore - Learning continuous action models in a real-time strategy environment - 2008Book: AI Game Engine Programming - 2009
- 29. Thanks