1. CSE 473 Artificial Intelligence Review

2. Logistics Project reports and homework 5 due Monday 5pm Monday 5pm project demo in 002 Exam next Wed 8:30—10:30 Regular classroom Closed book Cover all quarter’s material Emphasis on material not covered on midterm

3. Defining AI thought vs. behavior human-like vs. rational Systems that think like humans Systems that think rationally Systems that act like humans Systems that act rationally

4. Goals of this Course To introduce you to a set of key: Paradigms & Techniques Teach you to identify when & how to use Heuristic search Constraint satisfaction Logical inference Bayesian inference Policy construction Machine learning

5. Theme I Problem Spaces & Search How to specify PS? Two kinds of search?

6. Learning as Search Decision trees Structure learning in Bayesian networks Unsupervised clustering

7. Theme II In the knowledge lies the power Adding knowledge to search

8. Heuristics How to generate? Admissibility?

9. Propositional. Logic vs. First Order Ontology Syntax Semantics Inference Algorithm Complexity Objects, Properties, Relations Atomic sentences Connectives Variables & quantification Sentences have structure: terms father-of(mother-of(X))) Unification Forward, Backward chaining Prolog, theorem proving DPLL, GSAT Fast in practice Semi-decidableNP-Complete Facts (P, Q) Interpretations (Much more complicated) Truth Tables

10. © D. Fox 10 Planning  Problem solving algorithms that operate on explicit propositional representations of states and actions.  Make use of specific heuristics.  State-space search: forward (progression) / backward (regression) search  Partial order planners search space of plans from goal to start, adding actions to achieve goals  GraphPlan: Generates planning graph to guide backwards search for plan  SATplan: Converts planning problem into propositional axioms. Uses SAT solver to find plan.

11. Probabilistic Representations How encode knowledge here? In the knowledge lies the power

12. Uncertainity Joint Distribution Prior & Conditional Probability Bayes Rule [Conditional] Independence Bayes Net Propositional Hot topic: extensions to FOL Earthquake Burglary Alarm Nbr2CallsNbr1Calls Pr(B=t) Pr(B=f) 0.05 0.95 Pr(A|E,B) e,b 0.9 (0.1) e,b 0.2 (0.8) e,b 0.85 (0.15) e,b 0.01 (0.99) Radio

13. © D. Fox 13 P(B | J=true, M=true) EarthquakeBurglary Alarm MaryJohn P(b|j,m) =   P(b,j,m,e,a) e,a

14. © D. Fox 14 P(B | J=true, M=true) EarthquakeBurglary Alarm MaryJohn P(b|j,m) =  P(b)  P(e)  P(a|b,e)P(j|a)P(m,a) e a

15. Localization as Dynamic Bayes Net m x z u x z u 2 2 x z u... t t x 1 1 0 10t-1

16. Markov Assumption Helps! x t-1 x z u t t t

17. © D. Fox 17

18. Representations for Bayesian Robot Localization Discrete approaches (’95) Topological representation (’95) uncertainty handling (POMDPs) occas. global localization, recovery Grid-based, metric representation (’96) global localization, recovery Multi-hypothesis (’00) multiple Kalman filters global localization, recovery Particle filters (’99) sample-based representation global localization, recovery Kalman filters (late-80s?) Gaussians approximately linear models position tracking AI Robotics

19. Sensor board: Data Stream Courtesy G. Borriello

20. Example Evaluation Run Decision stumps classifiers (at 4Hz) HMM with probabilities as inputs (using a 15 second sliding window with 5 second overlap) Ground truth for a continuous hour and half segment of data.

21. Specifying an MDP S = set of states set (|S| = n) A = set of actions (|A| = m) Pr = transition function Pr(s,a,s’) represented by set of m n x n stochastic matrices each defines a distribution over SxS R(s) = bounded, real-valued reward fun represented by an n-vector

22. Max Bellman Backup, Value Iteration a1a1 a2a2 a3a3 s VnVn VnVn VnVn VnVn VnVn VnVn VnVn Q n+1 (s,a) V n+1 (s)

23. © D. Fox 23 Stochastic, Fully Observable

24. © D. Fox 24 Why is Learning Possible? Experience alone never justifies any conclusion about any unseen instance. Learning occurs when PREJUDICE meets DATA!

25. © D. Fox 25 Inductive learning method Construct/adjust h to agree with f on training set (h is consistent if it agrees with f on all examples) E.g., curve fitting: Ockham’s razor: prefer the simplest hypothesis consistent with data

26. © D. Fox 26 Decision Tree Overfitting Number of Nodes in Decision tree Accuracy 0.9 0.8 0.7 0.6 On training data On test data

27. © D. Fox 27

28. © D. Fox 28

29. © D. Fox 29

30. © D. Fox 30

31. © D. Fox 31

32. © D. Fox 32

33. © D. Fox 33

34. © D. Fox 34

35. © D. Fox 35 Anemia Group Control Group

36. And More Specific search & CSP algorithms Adversary Search Inference in Propositional & FO Logic Lots of details