Intelligent Decision Support Systems: A Summary
Case-Based Reasoning Example: Slide Creation Repository of Presentations: -5/9/00: ONR review -8/20/00: EWCBR talk -4/25/01: DARPA review Specification Revised talk 3. Revise Slides of Talks w/ Similar Content 1. Retrieve 5. Retain New Case 4. Review New Slides - 9/12/03: cse395 First draft 2. Reuse cse395 E-commerce (Joe Souto) Recommender (Chad Hogg) Conversational CBR (Shruti Bhandari) MDPs and Reinforcement Learning (Megan Smith) Fuzzy Logic (Mark Strohmaier) 6 lectures + programming project E-commerce (Joe Souto) Recommender (Chad Hogg) Conversational CBR (Shruti Bhandari) MDPs and Reinforcement Learning (Megan Smith) Fuzzy Logic (Mark Strohmaier) 6 lectures + programming project Design (Liam Page) Rule-based Systems (Catie Welsh) Configuration (Sudhan Kanitkar) Intelligent Tutoring Systems (Nicolas Frantzen) 2 lectures Design (Liam Page) Rule-based Systems (Catie Welsh) Configuration (Sudhan Kanitkar) Intelligent Tutoring Systems (Nicolas Frantzen) 2 lectures Case Base Maintenance (Fabiana Prabhakar) Help-desk systems (Stephen Lee-Urban) 2 lectures (indexing) Case Base Maintenance (Fabiana Prabhakar) Help-desk systems (Stephen Lee-Urban) 2 lectures (indexing)
Knowledge Representation (Prof. Jeff Heflin) Inferred Hierarchy DL Reasoner Ontology table & view creation Database operation Rule-Based Systems (Catie Welsh) Rule inference as search trees Advantages: volume of information, prevent mistakes Disadvantages: lack of flexibility to changes in environment Real world domain: IDSS for cancer test
Design (Liam Page) Constrains not fully specified (ranking by preference) Graph representation of data Flexible similarity metrics: local Model+cases Fish and Shrink retrieval Configuration Systems (Sudhan Kanitkar) Concept Hierarchies Structure-Based Approach Forms of adaptation: Compositional Transformational
E-commerce (Joe Souto)Recommender Systems (Chad Hogg) fixed innovative products Knowledge gap: seller doesn’t know what buyer wants User Requirements Hard versus soft Redundant + contradictory Local similarity metrics Information overload Variants: Content: inter-item similarity Collaborative: Preferences Query based Hybrid Compromise-driven retrieval
Intelligent Tutoring Systems (Nicolas Frantzen) Description/performance history of student behavior Information the tutor is teaching Reflects the differing needs of each student Help-desk systems (Stephen Lee-Urban) Experience Management CBR Approved versus Open cases Client-Server architecture But all share domain model Help-desk deployment processes: Technical: requirements Organizational: training Managerial: quality assurance
Conversational Case-Based Reasoning (Shruti Bhandari) Case Base Maintenance (Fabiana Prabhakar) Contrast with rule-based systems Initial input in plain text Only relevant cases/questions shown to user Coverage(CB): all problems that can be solved with CB Reachability(P): all cases that can solve P
MDPs and Reinforcement Learning (Megan Smith) Fuzzy Logic (Mark Strohmaier) Policy : state action MDPs: probabilities are given RL: learn the probabilities (adaptive) Drops concept of an element either belongs to a set or not Rather there is a degree of membership As a result well capable of dealing with noise Applications: autonomous vehicles
TopicPresenterKnowledgeCertaintyTask OntologiesProf. Heflin Intensive Certain Methodological Rule-Based Systems Catie Welsh IntensiveUncertaintyAnalysis DesignLiam PageIntensiveCertainSynthesis ConfigurationSudhanKanitkarIntensiveCertainSynthesis E-commerceJoe SoutoLow/MediumUncertaintyAnalysis RecommenderChad HoggLow/MediumUncertaintyAnalysis Intelligent Tutor. Systems Nicolas Frantzen IntensiveCertainAnalysis/ Synthesis Help-desk systems Stephen Lee- Urban Low/MediumUncertaintyAnalysis CCBRShruti BhandariLow/MediumUncertaintyAnalysis CBMFabiana Prabhakar LowN.A.Methodological MDPs and RLMegan SmithLow/MediumUncertaintyMethodological Fuzzy LogicMark Strohmaier MediumUncertaintyMethodological
Computational Complexity Techniques for IDSS have a variety of complexities Searching for m-NN in a sequential case base with n cases: O(nlog 2 m) Searching for m-NN in a case base with n cases indexed with a KD-tree : O(log k n log 2 m) Constructing optimal decision tree, graph-subraph isomorphism, configuration, planning, constraint satisfaction NP-complete Quantified Boolean formulas, hierarchical planning, winning strategies in games PSPACE-complete
Computational Complexity Programming project Applications to IDSS: Analysis Tasks Help-desk systems Classification Diagnosis Tutoring Synthesis Tasks Int. Tutoring Systems E-commerce Help-desk systems AI Introduction Overview IDT Attribute-Value Rep. Decision Trees Induction CBR Introduction Representation Similarity Retrieval Adaptation Rule-based Inference Rule-based Systems Expert Systems The Summary Synthesis Tasks Constraints Configuration Uncertainty (MDPs, Fuzzy logic)