10-706 Graphical, Statistical and Causal Models

Units: 12.0 Department: Center for Auto. Learning & Disc. Cross-listed: 80-512 , 80-812 Related URLs: http://www.cmu.edu Statisticians and computer scientists have developed a set of statistical and causal models that incorporate the use of various kinds of graphs as a key element in the model structure. I will refer to this set of causal and statistical models simply as graphical models. The goal of this course is to teach students how to construct graphical models from data and various background assumptions, the various uses of graphical models, the assumptions that underlie the methods of construction and the uses of the models, and the relative advantages and disadvantages of graphical models compared to other kinds of statistical and causal models. The two main uses of graphical models are estimating and calculating conditional probabilities, and estimating and calculating manipulated probabilities (the probability distribution in a population that results when the value of a variable has been set, as in a randomized experiment). Estimating a conditional probability distribution is the appropriate operation when performing classification. Estimating a manipulated probability is the appropriate operation, for example, when trying to guess the consequences of an action, or when comparing different plans. There are three main steps in using graphical models for estimating conditional probabilities or manipulated probabilities: construct the graph; estimate the parameters of the graph; and use the estimated parameters and the graph to estimate the conditional probabilities or manipulated probabilities. For each of these steps, I will describe the variety of different kinds of background assumptions that can be made, and both Bayesian and frequentist methods for carrying out the steps.

Popularity index Rank for this semester: #0 Rank in this department: #0   Students also scheduled 21-690 Methods of Optimization

Spring 2005 times

No sections available for semester Spring 2005.