DCS 530 -- Principles of Artificial Intelligence

Fall 2001

Main Page Contents
About this course
Schedule
Lectures
Materials
Links

Course People
Matthew Stone
Rong Zhang

Schedule

Class Tuesday/Thursday 4:30-5:50, Hill 254

Announcements

Dec 16
Reminder: the exam is either at 9am or 4pm December 18 in Hill 254.
Caveat: there is a typo in the justification of the ANOVA in Cohen, on page 193. The between-group variance should include an additional factor Nj - the number of data points in group j.
Good luck.
Nov 27
Short Written Exercises
Due December 6.
Nov 13
Final paper - a short proposal
Due December 11
(Plenty of time to think about it.)
Reminder: No office hours Nov 14, Dimitris Metaxas Nov 15.
Nov 8
Written Exercises
Due November 20
Updated and synchronized syllabus on this page.
Oct 4
Correction to Homework One
Small typo fixed in mystery algorithm of extra credit A
Oct 3
Written Exercises
Due October 18
Sep 27
Updated and synchronized syllabus on this page.
Sep 25
Homework One
Due October 11
Aug 29
Who should take this class?

Lecture Schedule, AI Events, Notes

Sep 4
What is AI?
Module 1: A Prototypical Case Study in AI
Reading: Agents in the Real World
Homework: Decision Analysis
Lectures: Sep 6-Sep 20
Sep 6
Decision analysis as a computational model (1)
Actions, observations, outcomes; probability and utility.
Sep 11
Sep 13
Decision analysis as a computational model (2)
Interpreting models; solving for and carrying out policies in agents.
Sep 18
Decision analysis as a computational model (3)
Design and evaluation; sensitivity analysis, statistical hypotheses about running agents.
Sep 20
Decision analysis as a computational model (3)
Computational complexity; training data; model estimation and model induction; generalization, model selection, data sparsity.
Module 2: Perception and Bayesian Inference
Reading: Chris Bishop, Neural Networks for Pattern Recognition Chapters 1-3
Homework: Written Exercises
Lectures: Sep 25-Oct 11.
Sep 25
Bayesian analysis and models for classification (1)
Motivation for a Bayesian approach. (Ch 1.1-1.3; 1.8)
Sep 27
Bayesian analysis and models for classification (2)
Naive Bayes inference (discrete case). Text classification. (Ch 3.1.4)
Oct 2
Bayesian analysis and models for classification (3)
Continuous variables, normal distributions, linear classifiers. (Ch 1.8; Ch 3.1; Ch 3.2)
Oct 4
Bayesian analysis and models for classification (4)
Learning from training data. Maximum likelihood. (Ch 1.9 and 2.1-2.4)
Oct 9 Guest Lecture: Haym Hirsh
Current research in text classification.
Oct 11 Bayesian analysis and models for classification (5)
General density estimation: nearest neighbor classification. (Ch 2.5 and 2.5)
Oct 16
Bayesian analysis and models for classification (5)
Clustering; k-means; expectation maximization.
Module 3: Time
Reading: Eugene Charniak, Statistical Language Learning Chapters 1-7
Maybeck. Introduction, from Stochastic Models, Estimation and Control
Russell and Norvig. Chapter 15, Probabilistic Inference, from Artificial Intelligence, a Modern Approach
Homework: Written Exercises
Lectures: Oct 18-Nov 15.
Oct 18
Models of time (1)
Markov models and hidden Markov models.
Charniak Chapter 2 and 3.1-3.2.
Oct 23
Models of time (2)
HMM decoding. Part-of-speech tagging.
Charniak Chapter 3.3 and Chapter 4.1.
Oct 25
Midterm.
Oct 30
Models of time (3)
HMM inference. Forward/backward.
Charniak Chapter 4.2
Nov 1
HMM inference: Training. Speech and gesture recognition.
Graphical representations of probabilistic models.
Charniak Chapter 4.3; Russell and Norvig Chapter 15.
Nov 6
Models of time (4)
The Kalman filter. Tracking and learning with Gaussian priors.
Maybeck, Introduction.
Nov 8
Models of hierarchical structure. Trees, CFGs and PCFGs.
Charniak Chapter 5.
Nov 13
Algorithms for PCFGs.
Charniak Chapter 6.
Nov 15
Guest Lecture: Dimitris Metaxas.
Current research in visual tracking and recognition.
Module 4: Planning
Reading: Russell and Norvig. Chapters 16, 17 and 20, from Artificial Intelligence, a Modern Approach
Nov 20
General probabilistic inference: influence diagrams.
Nov 27
Markov decision processes: Value iteration.
Nov 29
Markov decision processes: Policy iteration.
Module 5: Evaluation
Reading: Cohen. Chapters 3 and 6, from Empirical Methods in Artificial Intelligence
Dec 4
Evaluation (1).
Pitfalls and methodology.
Dec 6
Evaluation (2).
Performance metrics.
Dec 11
Evaluation (3).
Training and test data. Reliability. Cross-validation.
Dec 18
4-7 pm, Hill 254: Final.

Materials

Texts
Neural networks for pattern recognition. Christopher Bishop. Oxford, 1995.
Statistical language learning. Eugene Charniak. MIT, 1993.
On Reserve in the Math Library
Artificial Intelligence: A Modern Approach. Stuart Russell and Peter Norvig, Prentice Hall, 1995. Chapters 15, 16, 17 and 20.
Empirical Methods for Artificial Intelligence. Paul Cohen, MIT, 1995. Chapters 3 and 6.
Notes
Agents in the Real World: Computational Models in Artificial Intelligence and Cognitive Science. Matthew Stone. Revised version of this chapter will appear in Zenon Pylyshyn and Ernie Lepore, eds., What is cogntive science, (second edition) Blackwell, 2002.
Peter Maybeck's introduction to the Kalman filter, a reference for the material covered in class October 30. Source: Maybeck, 1979 Stochastic Models, Estimation and Control, Chapter 1, "Introduction", pp 1-15. Part of a general web resource on the Kalman filter.
Research Articles (following up class material, for people who are curious)
The interactive museum tour-guide robot. Burgard et al, AAAI (National Conference on Artificial Intelligence) 1998. An overview of a state-of-the-art agent.
Hierarchically classifying documents using very few words. Koller and Sahami, ICML (International Conference on Machine Learning) 1997. A description of a principled and effective probabilistic text classifier.
Real-Time American Sign Language recognition from video using hidden Markov models Thad Starner and Alex Pentland. An illustration of the breadth of HMM techniques for AI---recognizing a visual language.
CONDENSATION - conditional density propagation for visual tracking Michael Isard and Andrew Blake, International Journal on Computer Vision, 1998. This paper uses sampling to represent probability density for a computer vision application that has to deal with visual ambiguity.
Statistical parsing with a context-free grammar and word statistics Eugene Charniak, AAAI (National Conference on Artificial Intelligence) 1997. An important update to the book, that shows how to build a PCFG grammar for English that disambiguates reliably.
Packet Routing in Dynamically Changing Networks: A Reinforcement Learning Approach Justin Boyan and Michael Littman. NIPS (Neural Information Processing Systems Conference), 1994.
An application of reinforcement learning to dialogue strategy selection in a spoken dialogue system for email. Marilyn A. Walker. Journal of Artificial Intelligence Research, Volume 12, pages 387-416, 2000. An illustration of connections between MDPs, reinforcement learning, and performance evaluation in agents.

Links