• Boaz Lerner
  Learning Probabilistic Graphical Models
• Dan Steinberg Ph.D.
  Data Mining: CART / MARS

Abstract
The tutorial will focus on inference and learning issues in probabilistic graphical models. It will distinguish between directed and undirected graphical models, where the former are usually called Bayesian (or Belief) networks and the latter Markov random fields. Supervised and unsupervised learning of the structure and parameters of Bayesian networks will be demonstrated. Since inference and learning in graphical models representing real-world problems are often found to be intractable, different approximations, such as Laplace, Monte Carlo and variational, will be suggested. The framework of variational methods will be described and the application of the methods to inference and learning in intractable graphical models will be presented.

Abstract

• CART¨ (Classification and Regression Trees)
  CART¨ is the only decision-tree tool based on the original CART code developed by Stanford and UC Berkeley statisticians. Offering superior speed and ease-of-use, CART automatically provides insight into data and produces highly accurate, intelligible predictive models.
• MARSú (Multivariate Adaptive Regression Splines)
  MARSú, is a high-speed flexible regression tool for data mining and predictive modeling that is a natural alternative to neural networks and more conventional regression models.

Dan Steinberg has over 20 years of experience in data mining and statistical consultation. Steinberg, who received his Ph.D. in Econometrics from Harvard University, has served on the technical staff (MTS) at AT&T Bell Laboratories, as Assistant Professor of Economics at the University of California, San Diego, and as a consultant for numerous Fortune 100 clients. In addition to publishing articles in statistical, econometric, computer science, and marketing journals, he has developed a series of advanced statistical-analysis programs for Salford Systems. Steinberg has received awards from the SAS Usersâ Group International and the American Marketing Association and has been a featured speaker on data-mining issues for the American Marketing Association, the American Statistical Association, the Direct Marketing Association and at DCIâs Database and Client/Server Conference. A book on CART¨ by Dan Steinberg, Yuji Horie and Atsushi Ootaki was just awarded Japanâs Nikkei Quality Control Literature Prize by the Deming Prize Committee.