CSE 574 - Artificial Intelligence II

University of Washington Department of Computer Science & Engineering

CSE 574 - Artificial Intelligence II - Winter 2000

CSE Home

About Us

Contact Info

Data Mining and Machine Learning

Instructor: Pedro Domingos
Office: Sieg 216
Office hours: Thursday 11:00-12:00 and by appointment

TA: Geoff Hulten
Office: 104 Chateau
Office hours: ?? and by appointment

Class meets: TuTh 12:00-1:20 EE1 026

Topics

Introduction
Classification
- Decision trees
- Rule induction
- Nearest neighbor
Model ensembles
Regression
Probability estimation
Clustering
Association rule mining
Scaling up learning algorithms
Data preparation for data mining
Postprocessing learned models

Assignments

Students will do a project and give a seminar. Projects can be done individually or in groups of two, and are due on the last day of classes (March 9). Projects can be chosen from the list below, or proposed by the students. The seminar may be on a topic related to the project, or not. Seminars are Pass/Fail; a Pass is required to complete the class. The class grade will be the project grade.

Textbooks

Tom Mitchell, Machine Learning, McGraw-Hill, 1997.
Usama Fayyad, Gregory Piatetsky-Shapiro, Padhraic Smyth, and Ramasamy Uthurusamy (editors), Advances in Knowledge Discovery and Data Mining, AAAI Press/MIT Press, 1996.
David Hand, Heikki Mannila, and Padhraic Smyth, Principles of Data Mining, MIT Press, to appear.

Papers

T. Dietterich, Machine-learning research: Four current directions, AI Magazine, 1997.
P. Domingos, Machine learning, Handbook of Data Mining and Knowledge Discovery, Oxford UP, to appear.
G. H. John, Behind-the-scenes data mining, SIGKDD Explorations, 1999.

Readings

Week 1: Chapter 1 of Fayyad, Chapter 1 of Mitchell, paper by John
Week 2: Paper by Domingos, Chapter 3 of Mitchell
Weeks 3 & 4: Chapter 10 of Mitchell, Chapter 5 of Fayyad
Week 5: Chapter 8 of Mitchell, Section 2 of paper by Dietterich
Week 6: Chapter 4 of Mitchell
Week 7: Chapter 6 of Mitchell, Chapter 11 of Fayyad

Lecture Notes

Week 1: Introduction
Week 2: Decision trees: Part 1 and Part 2
Weeks 3 & 4: Rule induction
Week 5: Nearest neighbor, model ensembles
Week 6: Neural networks: Part 1 and Part 2
Week 7: Bayesian networks: Part 1 and Part 2

Seminars

Tu Feb 22: Sarah & Steve (data cleaning)
Th Feb 24: Richard (pruning adaptive boosting, knowledge discovery via multiple models) & Ana-Maria (learning with concept drift and hidden context)
Tu Feb 29: Corin (intrusion detection) & Mike (support vector machines)
Th Mar 2: Matt (hidden Markov models), Vassili & Alex (discovering frequent episodes in sequences)
Tu Mar 7: Ashish & Deepak (causal discovery)
Th Mar 9: AnHai (text classification) & Yongshao (error-correcting output codes)

Project Ideas

Extending the scope of learning algorithms
- Learning time-changing concepts
- Learning given a query distribution
- Learning with thousands of classes
- Learning from non-random samples
- Learning from multiple databases
- Learning from mixed-media data
- Learning across populations
Web mining
- Modeling the hidden Web through querying
- Learning to place Web pages in Yahoo! categories
- Clickstream mining: From sequences to shortcuts
- Web dynamics: Predicting changes in access frequency
Scaling up
- Adaptive sampling
- Scalable data squashing
- Interleaving learning and compression
Using frequent sets for prediction
- Extending naive Bayes with frequent sets
- Frequent sets as features for large-margin classifiers
- Probability distributions from frequent sets via iterative scaling
- Learning Bayes nets via frequent sets
Overfitting avoidance
- Estimating the level of noise in a database
- Learning noise models
- Empirical process-oriented evaluation of decision trees
- Knowledge-intensive evaluation functions
Model ensembles
- Combining multiple decision trees into one
- Avoiding overfitting in boosting
- Scaling up alternating decision trees
Learning with prior knowledge
- Learning with weak causal knowledge
- Learning to correct input errors
- Online handwriting recognition

Mail Archive

Department of Computer Science & Engineering
University of Washington
Box 352350
Seattle, WA 98195-2350
(206) 543-1695 voice, (206) 543-2969 FAX
[comments to ghulten@cs.washington.edu]