Data Mining and Machine Learning

General Information
Course Contents
Application
Schedule

General Information

SAMSI is instituting a new summer activity—technology transfer short courses designed to consolidate results from (in most cases, earlier years') SAMSI programs, and to make the results available to working professionals in a compact, hands-on format. The first such course is derived from the 2003-04 SAMSI program on Data Mining and Machine Learning (DMML).

The goals of the DMML technology transfer short course are to:

• Provide a survey of the theoretical basis for modern data mining


• Give participants hands-on experience with data mining software


• Convey insights and strategies for data mining practice.

The theoretical component will emphasize ideas over rigor; the software component will sample the major techniques that are now commonly used for visualization, classification, and regression; and the applications component will walk participants through the practical analysis of some famous real-world data sets.

The structure of the short course is that there will be three hours of lecture each morning. Each afternoon will start with a 90 minute computer lab that goes over an application using real data and relevant software, followed by a 90 minute lecture by a guest speaker. There will be several breaks during the day.

The course begins with an introductory overview of data mining: its scope, classical approaches, and the heuristics that guided the initial development of theory and methods. Then the course moves towards the treatment of more modern issues such as boosting, overcompleteness, and large-p small-n problems. This leads to a survey of currently popular techniques, including random forests, support vector machines, wavelets, and PAC bounds.

The main focus is upon a central focus of the SAMSI DMML program—regression inference, a paradigm that informs many data mining applications, but we also discuss clustering, classification, and multidimensional scaling.

The prerequisites for the course are a basic knowledge of applied multivariate inference and a general level of statistical knowledge comparable to a master's degree. Any math will focus upon conveying general insight rather than specific details.

Course Contents

1. Background and Overview: Nonparametric Regression, Cross-Validation, the Bootstrap


2. Key Ideas and Methods: Smoothing, Bias-Variance Tradeoff


3. Search and Variable Selection: Experimental Design, Gray Codes, Fitness


4. Nonparametric Regression: Heuristics on Eight Methods


5. Comparing Methods: Designing Experiments in Data Mining


6. Local Dimension: How to Pick Problems Wisely


7. Classification: Boosting, Random Forests, Support Vector Machines


8. Cluster Analysis: Hierarchical, k-Means, and Mixture Models; SOM


9. Issues with Bases: Hilbert Space, Shrinkage, Overcompleteness


10. Wavelets: Introduction, Construction, Examples


11. Structure Extraction: Regression and Multidimensional Scaling


12. Vapnik-Cervonenkis Classes and PAC Bounds

Principal instructor for the course will be David L. Banks, Professor of the Practice of Statistics and Decision Sciences at Duke University, and co-leader of the SAMSI DMML program.

Application

REGISTRATION IS NOW CLOSED

Schedule

Entire site © 2001-2008, Statistical and Applied Mathematical Sciences Institute. All Rights Reserved.