DCS 530 -- Principles of Artificial Intelligence

Course People
Matthew Stone
Aynur Dayanik

Class
Tuesday/Thursday 4:30-5:50, Core 301A

Office Hours
MS: Thursday 10-12 (Core 328).
AD: Tuesday 2-4 (Core 333).

News

• Tuesday, Dec 16, 4:00 pm
  Final, Hill 254.
  The exam will be 90 minutes.
  Those with another 6pm exam may begin at 3:15pm.
  

• Dec 11
  Selected answers to sample problems (no pictures).

• Dec 4
  Sample problems.

• Nov 20
  Additional readings will be available presently in the Math library and on electronic reserve.

• Nov 13
  Homework 3 corrected. Numerical change will make answering 1h easier.

• Nov 6
  Homework 3 out, due Nov 18.
  Final paper out, due Dec 9.

• The midterm, Oct 21
  will take place back in Hill 254.

• Oct 2
  Homework 2 out, due Oct 14.

• Sep 30
  A reminder to hand in your program in homework 1 electronically by email to mdstone@cs.

• Sep 11
  Quick clarification on correlation, as used by Paskin to describe the SLAM problem. Correlation can be understood intuitively as the interdependence of variable quantities. It is measured in terms of covariance.

  ...it is difficult to employ the covariance as an absolute measure of dependence because its value depends on the scale of measurement and so it is hard to determine whether a particular covariance is large at first glance. This problem can be elimnated by standardizing its value, using the simple coefficient of linear correlation. The population linear coefficient of correlation, ρ, is related to the covariance and is defined as ρ = covariance(X,Y)/(σ(X)σ(Y)) where σ(X) and σ(Y) are the standard deviations of X and Y respectively. [Mendenhall et al, Math. Stats. with Applications, PWS Kent 1990]

• Sep 09
  Reminder: Class is moved to Core 301A.
  Homework One out.

• Sep 02
  Initial revision of this web page.
  Announcement: Who should take this class?

Lecture Schedule, AI Events, Notes

• Sep 02
  What is AI?

• Sep 04
  Research practice and ideas in AI.
  Readings:
  Mark Paskin. Thin Junction Tree Filters for Simultaneous Localization and Mapping. IJCAI 2003, pages 1157-1164.
  Darse Billings, Neil Burch, Aaron Davidson, Robert Holte, Jonathan Schaeffer, Terence Schauenberg and Duane Szafron. Approximating Game-theoretic Optimal Strategies for Full-scale Poker. IJCAI 2003, pages 661-668.

• Module 1: A Prototypical Case Study in AI
  Reading: Agents in the Real World
  Supplementary material: Clemen, Decision Analysis, Chapters 1-5.
  Available on reserve.
  Homework: Decision Analysis due Sep 30
  Lectures: Sep 09-Sep 18

• Sep 09
  Decision analysis as a computational model (1)
  Actions, observations, outcomes; probability and utility.

• Sep 11
  Decision analysis as a computational model (2)
  Interpreting models; solving for and carrying out policies in agents.

• Sep 16
  Decision analysis as a computational model (3)
  Design and evaluation; sensitivity analysis, statistical hypotheses about running agents.

• Sep 18
  Decision analysis as a computational model (4)
  Computational complexity; training data; model estimation and model induction; generalization, model selection, data sparsity.

• Module 2: Perception and Bayesian Inference
  Lectures: Sep 23-Oct 14.
  Resources: Andrew Moore's computational statistics tutorials.
  Supplementary material: Duda, Hart and Stork, Pattern Classification, Chapters 1-3.
  Available on reserve.

• Sep 23
  Bayesian analysis and models for classification (1)
  Motivation for a Bayesian approach.
  Resources: Probabilistic and Bayesian analytics.
  

• Sep 25
  Bayesian analysis and models for classification (2)
  Naive Bayes models (discrete case). Text classification.

• Sep 30
  Bayesian analysis and models for classification (3)
  Classifying with Naive Bayes models; feature space and linear classifiers.

• Oct 02
  Bayesian analysis and models for classification (4)
  Training and Naive Bayes models.
  Homework: Model assumptions: due Oct 14
  

• Oct 07
  Bayesian analysis and models for classification (5)
  Continuous variables, normal distributions, linear classifiers.
  Resources: Probability Density Functions.
  Resources: Gaussians.

• Oct 09
  Bayesian analysis and models for classification (6)
  Multivariate Gaussians. Classifying with Gaussians.
  Resources: Gaussian Bayes Classifiers.

• Oct 14
  Bayesian analysis and models for classification (7)
  Maximum likelihood estimation. Density estimation. Clustering.
  Resources: Gaussian Mixture Models.

• Oct 16
  Review.

• Oct 21
  Midterm.

• Module 3: Time
  Lectures: Oct 23-Nov 06.

  Resources: Andrew Moore's Tutorial Slides on Hidden Markov Models
  Larry Rabiner's tutorial article on HMMs. (Laurence R. Rabiner, A tutorial on Hidden Markov Models and Selected Applications in Speech Recognition. Proceedings of the IEEE 77(2):257-286, 1989).
  Greg Welch and Gary Bishop's web resource on the Kalman filter.
  Peter Maybeck's introduction to the Kalman filter. Source: Maybeck, 1979 Stochastic Models, Estimation and Control, Chapter 1, "Introduction", pp 1-15.

• Oct 23
  Models of time (1)
  Markov models and hidden Markov models.
  

• Oct 28
  Models of time (2)
  HMM inference. Forward/backward.
  

• Oct 30
  Models of time (3)
  HMM decoding. Part-of-speech tagging.
  

• Nov 04
  HMM inference: Training. Speech and gesture recognition.
  Graphical representations of probabilistic models.
  

• Nov 06
  Models of time (4)
  The Kalman filter. Tracking and learning with Gaussian priors.
  

• Module 4: Planning
  

• Nov 11
  General probabilistic inference: belief networks.
  Reading: Russell and Norvig, Probabilistic Reasoning (chapter 15 of Artificial Intelligence).
  Charniak, Bayesian Networks without Tears, from AI Magazine, 1991.
  Tutorial Slides on Bayesian networks.

• Nov 13
  General probabilistic inference: influence diagrams.
  Reading: Russell and Norvig, Making simple decisions (chapter 16 of Artificial Intelligence).
  Supplementary reading: Clemen, Decision Analysis, Chapters 3 and 4.
  

• Nov 18
  Markov decision processes: Value iteration.
  Reading: Russell and Norvig, Making complex decisions (chapter 17 of Artificial Intelligence).
  Tutorial Slides on Markov decision processes.

• Nov 20
  Markov decision processes: Policy iteration; reinforcement learning.
  Reading: Russell and Norvig, Making complex decisions and Reinforcement learning (chapters 17 and 20 of Artificial Intelligence).
  Supplementary reading: Howard, Dynamic Programming and Markov Processes, excerpts, especially Chapter 7.
  Tutorial Slides on Reinforcement learning.

• Nov 25
  Games, strategies, and coordination.
  Tutorial Slides on game theory.

• Module 5: Evaluation
  

• Dec 02
  Evaluation (1).
  Pitfalls and methodology.
  Reading: Cohen, Empirical Methods for AI, Chapter 3.

• Dec 04
  Evaluation (2).
  Performance metrics.
  Reading: Cohen, Empirical Methods for AI, Chapter 6.

• Dec 09
  Evaluation (3).
  Training and test data. Reliability. Cross-validation.

• Tuesday, Dec 16, 4:00 pm
  Final.

Materials
This class will work from an eclectic combination of survey and tutorial papers, research articles, course notes, and other resources. This list will grow as the semester proceeds.

• Articles
  
  • Mark Paskin. Thin Junction Tree Filters for Simultaneous Localization and Mapping. IJCAI 2003, pages 1157-1164.
    

  • Darse Billings, Neil Burch, Aaron Davidson, Robert Holte, Jonathan Schaeffer, Terence Schauenberg and Duane Szafron. Approximating Game-theoretic Optimal Strategies for Full-scale Poker. IJCAI 2003, pages 661-668.
    

• 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, 2004.

Links

• General AI References
  • Hal's Legacy: 2001's Computer as Dream and Reality. An electronic book surveying how Kubrick and Clarke's hypothetical computer compares with the actual state of the art today.
  • The CMU Artificial Intelligence Repository collects files, programs and publications related to AI.
  • AAAI, the society for AI research has put together a web site about AI, including descriptions, references and web links.

• Cool AI Systems
  • AI technology really is being used onboard spacecraft! NASA's Deep Space 1 probe was operated by knowledge-based agent software during a test last fall.
    The algorithms and representations of this software involve industrial-strength implementations of the techniques taught in this course.
  • Researchers at NASA and the Robotics Institute at CMU have constructed an autonomous robot to find and classify meteors and other rocks in Antarctica.
  • An IBM team is using logic to guide the action of internet agents that run e-businesses.

• Robots and the Media
  • Fast, Cheap and Out of Control: The eerie documentary featuring Rod Brooks.
  • Buy Sony's entertainment robot dog.
  • Killer Robot HQ. "Survival Research Labs is among the first civilians to develop and operate a multi-user, teleoperated firing system using free software deployed over the web."

• More AI related Rutgers stuff
  • The general colloquium series of the Center for Cognitive Science presents cutting-edge research on the computational study of the mind.
  • The series on Human and Computer Vision features weekly talks by researchers spanning both human and computer vision.