CS 731: Advanced methods in artificial intelligence, with biomedical applications (Fall 2009)

Instructor: David Page
page@biostat.wisc.edu
Office: 6743 Medical Sciences Center (corner of Charter and University)
Office Hours: Noon-1 Mondays and 2-3 Fridays
Office Phone: 265-6168
TA: Yancan Huang
hyc@cs.wisc.edu
Office: 3351 CS
Office Hours: 10am-Noon Wednesdays
Office Phone: 890-0033

News: Exam is scheduled for Fri, Dec 4, 2:30-4:30, in our regular room.
News: Projects are due by email (PDF format, please), by 11:59pm on Thurs, Dec 17. Please use "731 Project" as subject. If you write code, please deposit that in the Handin directory labeled Project.
Prerequisite: CS 540 or equivalent
Meeting Time and Location: 2:30-3:45 MWF, 1257 Computer Sciences
Note: We reserve Fridays for make-up classes, reviews and exam. If we meet on Friday, I will always announce it 2 weeks in advance.
Textbooks (Both Required):
- Koller, D. and Friedman, N. (2009). Probabilistic Graphical Models: Principles and Techniques. Available in the Bookstore or on Amazon.
- Nilsson, Ulf and Maluszynski, Jan. (2000). Logic, Programming and Prolog (2nd ed). Available as pdf from this link.
Archive of class e-mail

Course Overview

Many advances in artificial intelligence build on mathematical logic (hence the picture of Gottlob Frege at the left), probability and statistics (particularly Bayesian statistics -- hence the picture of Rev. Thomas Bayes at the right), and combinatorics. This course will cover many of the most important recent advances and will illustrate their use with biomedical applications. The course does not assume knowledge of medicine (nor of biology or chemistry beyond the general familiarity possessed by a graduate student in sciences or engineering). It does assume a knowledge of data structures, algorithms, and search paradigms, as can be obtained for example by successfully completing a Data Structures course and an Introduction to AI course.

Course Outline

Review of Propositional Logic, Satisfiability, GSAT, WSAT
Bayesian Network Fundamentals
- Probability Basics, KF Ch 1-2
- Bayes Nets Syntax and Semantics, KF Ch 3
- Conditional Indenpendence in Bayes Nets, KF Ch 3, 5
- Basics of Exact Inference, KF Ch 9
- Exact Inference by Junction/Join/Clique Trees, KF Ch 10, see also optionally Huang and Darwiche 1995 (see also Spiegelhalter et al. 1993 sections 1, 2, 3, and 6, and Jensen & Jensen 1994)
- Approximation and Markov Chain Monte Carlo, KF Ch 12 (Sections 1-3)
- Foundations of Markov Chain Monte Carlo Sampling, KF Ch 12 (Sections 1-2)
- Belief Propagation, see optionally KF Ch 11 (Section 3)
- Undirected Models, KF Ch 4, 9, 10, 12 (Sections 1-3)
- Medical Applications in Pathology and Radiology, see optionally KF Box 3.D (pages 67-68) and Burnside et al., 2009
Learning in Bayes Nets
- Overview of Learning, Learning CPTs, KF Ch 16-17, optionally see also Heckerman 1996 tutorial
- Learning tree-augmented naive Bayes nets, PDF also available here
- Learning Network Structure in Bayes Nets, with application to gene expression data, KF Ch 18, optionally see also Heckerman tutorial, Friedman et al. 1999 (Learning Network Structure), Friedman et al. 1999 (Using Bayesian Networks to Analyze Expression Data), Heckerman, Geiger and Chickering, 1995 (Learning Network Structure, includes BD metric).
Statistical Relational Learning and Multi-Relational Data Mining
- Introduction to Statistical Relational Learning I by Sriraam Natarajan.
- Introduction to Statistical Relational Learning II by Sriraam Natarajan.
- First-order logic syntax and semantics (Powerpoint Lecture): Nilsson and Maluszynski, Ch. 1.
- Definite clause logic and least Herbrand models (Powerpoint Lecture): Nilsson and Maluszynski, Ch. 2.
- First Order Predicate Calculus examples
- Resolution and SLD-Resolution. (Powerpoint Lecture, PDF): Nilsson and Maluszynski, Ch. 3.
- Logic and Relational Databases. (Powerpoint Lecture, PDF): Nilsson and Maluszynski, Ch. 6.
- Lists and Recursion (Sample Code from This and Subsequent Lecture): Nilsson and Maluszynski, Ch. 7.
- Here is an excellent ILP reference, somewhat older but publicly available.
- The most recent issue of SIGKDD Explorations is a special issue devoted to multi-relational data mining.
- Some examples of biological applications of ordinary feature vector learning: gene chips and other types of high-throughput biological data, single-nucleotide polymorphism data and molecular bioactivity data, details of a particular gene chip application.
- The ILP specification, subsumption, saturation, lattice definitions, Bottom-up and Top-down search. (Powerpoint Lecture, PDF).
- Application to Drug Design---Pharmacophore discovery: Finn, Muggleton, Page, and Srinivasan, Machine Learning, Vol 30. A demonstration of this work is available in the directory ~cs731-1/public/Pharm -- see the README file there for directions.
- Other techniques for ILP and Multi-Relational Data Mining, including Propositionalization Techniques: Dzeroski, SIGKDD Explorations, 2003
- Discovering Structural Alerts for Mutagenicity: Srinivasan, King, Muggleton, and Sternberg, Artificial Intelligence Journal, Vol 85
- Probabilistic Relational Models: Getoor et al., 2002.
- Markov Logic Networks: Richardson & Domingos, 2006, Poon & Domingos, 2006.
- View Learning in Statistical Relational Learning: Davis et al., 2005
Time and Planning in Bayes Nets (1 week)
- Dynamic Bayes Nets; see this ISMB-02 paper for an application to gene expression microarray data. (Also available in PDF.)
- Influence Diagrams: From Influence Diagrams to Junction Trees; Jensen, Jensen & Dittmer, 1994)
- Markov Decision Processes (MDPs) and Partially-Observable MDPs (POMDPs): Planning and Acting in Partially Observable Stochastic Domains; Kaelbling, Littman & Cassandra, 1997)
Exam: TBA. Sample Exams (to be updated):
- A recent exam (Spring02), in MS Word format
- Sample 2 (postscript)
- Sample 3 (postscript)
- Sample 4 (postscript)
- Please note that earlier exams are shorter because I used to give two exams per semester, before there was a class project; now I give only one. So the most recent exam is the most indicative of length and breadth of material for future exams.
Course Requirements
The grading for the course will be be based on:
- Homework Assignments (2-3 programming and 2-3 written assignments anticipated): 40%
- Exam: 35%
- Project: 25%
Homework Policy
The homework assignments (both written and programming) and project are to be done individually. You may communicate with other class members about the problem, but please do not seek or receive help from people not in the class, and please do not share answers or code. Homeworks (both written and programming) are due at the start of class on the assigned due date, and late homeworks will be penalized 10 points (out of 100) for each lecture that passes after the assigned due date. At the start of the course every student will be given 3 "free" days, each of which may be used to offset a 10-point late penalty. (Free days are non-transferable, and no credit will be given for unused free days.) Unless otherwise specified, the executable code for each programming assignment should be deposited in the Handin directory at /p/course/cs731-dpage/handin/userid/proj#, where "#" denotes the number of the assignment and "userid" denotes your user ID on the CS machines. The directory will be closed at the start of class on the due date, so please email the TA before this time if you intend to use one or more of your free days. Written assignments should be turned in at the classroom, prior to the lecture.
Written Assignments
Written Homework 1, Assigned Sep 28, Due Oct 7: Variable Elimination, Junction Trees and d-Separation.
Written Homework 2, Assigned Oct 21, Due Oct 28: Learning Parameters from Incomplete Data: EM and Gibbs.

Programming Assignments
All programming assignments must be completed in one of Java, C, C++ or Yap Prolog.
Homework 1, Assigned Sep 9, Due Sep 23: Comparison of Fast Satisfiability Algorithms. Write your own implementations of GSAT and WSAT. Create your own test sets and compare performance (correctness and run-time) on satisfiability tasks of 100, 200, and 300 variables. It is suggested (though not required) that you test on satisfiable CNF formulas with 3 literals per clause (3-SAT) and 4.3 times as many clauses as variables, since that is what we will test it on. Your code should read the CNF formula from a file called "cnf". This file should start with the number of variables on a single line. It should then read one clause from each line. A clause should be a comma-separated list of literals such as the following:
X2,~X3,X4
where "~" denotes NOT and the variables are all named by "X" followed by a number. Please deposit source code to your handin directory, and please turn in a 1-page write-up in class comparing GSAT and WSAT. (Or you may also deposit your write-up to the handin directory.)
Programming Assignment 2, Assigned Nov 16, Due Nov 30: Programming Exercises in Prolog.

Project
Projects should be proposed by November 2 (verbal or email communication is acceptable). Projects must be done individually. The basis for the project grade will be your written report, which must be turned in no later than the last day of final exams. The report should be in the style of a conference paper, providing an introduction/motivation, discussion of related work, a description of your work that is detailed enough that the work could be replicated, and a conclusion. The format of the description of your work will depend on the nature of your project. If it is an implementation, then the description should make clear the algorithm(s) implemented and provide experimental results. If it is an application project, the description should say which system was used, how the data (or any other materials used) were collected, what experimental methodology was employed, and some estimate of the quality of the experimental results (e.g. a 10-fold cross-validation accuracy estimate). If it is a theoretical project, then the project description should consist of detailed definitions, theorems, and proofs. An example of an outstanding project report is here (Word File). You may choose projects from any area of AI (even those not covered in the course), but the following are some suggestions related to topics in the course.
- Build a large Bayes Net application and develop the CPTs based in part on real-world data.
- Build a Bayes net learning algorithm. To be non-trivial this should include either (1) learning in the case of missing data using EM, Gibbs-sampling, or gradient-ascent (not discussed in class) or (2) learning/modification of the BN structure as well.
- Implement an influence diagram -- a Bayes Net that also suggests actions to take.
- Write a first-principles proof of one of the properties discussed in class that we did not take time to prove. Examples include (1) every clique graph has a junction tree or (2) the Metropolis-Hastings algorithm converges to the desired stationary distribution.
- Implement a Dynamic Bayes Net or Dynamic Influence Diagram.
- Implement a novel ILP system and apply it to a toy domain or a real-world data set (for example, the Predictive Toxicology Evaluation set).
Some Freely-Available Software Related to Course Topics
INDUCTIVE LOGIC PROGRAMMING SYSTEMS
BAYESIAN NETWORK SYSTEMS

CS 731: Advanced methods in artificial intelligence, with biomedical applications (Fall 2009)

Course Overview

Course Outline

Course Requirements

Homework Policy

Written Assignments

Programming Assignments

Project

Some Freely-Available Software Related to Course Topics