ICS-275A, Constraint Networks, Fall 2001

announcements | home work | chapters & readings | project

Classoom: PSCB 220 Days: Tuesday & Thursday Time: 3:30 - 4:50pm. Instructor: Rina Dechter Textbooks Book chapters of "Constraint Processing" Course Goals The purpose of this course is to familiarize students with the theory and techniques of constraint processing, using the constraint network model. This model offers a natural language for encoding world knowledge in areas such as scheduling, vision, diagnosis, prediction and design, and it facilitates many computational tasks relevant to these domains. The course will focus on techniques for constraint processing. It will cover search techniques, consistency algorithms and structure based techniques, and will focus on properties that facilitate efficient solutions. Extensions will be given into applications such as temporal reasoning, diagnosis, scheduling, and probabilistic Bayes networks. The constraint language Eclipse may be used for modelling and solving constraint problems. The topics covered will be taken from the following list. Outline The constraint network model, examples. Graph representations, Properties of binary networks: equivalence, the minimal and the projection networks. Approximation algorithms: local-consistency vs. global-consistency, arc and path-consistency, directional-consistency, adaptive-consistency, relational-consistency, bucket-elimination. Backtracking strategies: Look-ahead schemes: forward-checking, variable and value orderings, constraint propagation. The Davis-Putnam algorithms. Look-back schemes: backjumping, constraint learning. Combining search and inference; the cycle-cutset scheme. Stochastic local search algorithms: SLS, GSAT, WSAT Constraint-based tractable classes: row-convexity, tightness, looseness, implicational and functional constraints, Horn clauses. Constraint optimization. More topology-based concepts and algorithms: tree-clustering, time-space tradeoff. Temporal constraint networks: quantitative, qualitative, and integrated temporal networks. Constraint Logic Programming Grading Policy Homeworks and projects (60%), Final (40%). Assignments: There will be weekly homework-assignments, a project, a final. Syllabus: Subject to changes Week Topic Date   Week 1 Tu, Chapters 1,2: Introduction: examples and definitions of Constraint networks. Th: Holiday (Yom Kipur). 09-25 Week 2 Tu, (longer class) Chapter 2: Propertries of Binary networks. Chapter 3: Consistency enforcing algorithms, arc, path and i-consistency Th, Chapter 3: continued. 10-02 Week 3 Tu/Th,Chapter 4: Graph concepts (induced-width), Directional consistency, Adaptive-consistency 10-09 Week 4 Tu/Th: Chapter 5: Backtracking search, look-ahead methods 10-16 Week 5 Tu/Th, Chapter 6: Backtracking search, look-back methods 10-23 Week 6 Tu, Chapter 7: combining seach and inference, the cycle-cutset scheme Th: Chapter 8: Stochastic local search 10-30 Week 7 Tu/Th, Chapter 9: Advanced consistency methods; Relational consistency and bucket-elimination 11-06 Week 8 Tu/Th, Chapter 10: Tractable constraint languages 11-13 Week 9 Tu, Chapter 11: Constraint optimization Th: Thanksgiving 11-20 Week 10 Additional topics taken from: Tree-clustering (Chapter 12), Temporal constraints (Chapter 13), Constraint Programming (Chapter 14) Final 11-27 Resources on the Internet Analytica (an alternative to spreadsheets) AI on the web The constraint language Eclipse On-line documentation CSPlib of benchmark CSPs "Bucket Elimination: A unifying framework for structure-driven inference," R. Dechter. "Bucket Tutorial."