Integer Programming Methods to Learn Causal Structures

3/20/25 | 4:15pm | E51-376

Sanjeeb Dash

Sanjeeb Dash

Researcher
IBM Research

Abstract: The problem of finding score-maximizing Bayesian Networks, where the score represents quality of fit to input data, can be modeled as an integer program, and some of the state-of-the-art algorithms for this problem solve such integer programs. We discuss recent work on integer programming models for important variants of the Bayesian Network Structure Learning problem that involve latent variables and focus on the problem of learning linear structural equation models.

Bio: Dr. Sanjeeb Dash is a member of the Mathematics and Theoretical Computer Science department of IBM Research and leads the Foundations of Optimization and Probability group located at IBM’s T. J. Watson Research Center. He works on both theoretical and practical aspects of Discrete Optimization. The focus of his research is Integer Programming and Linear Programming including applications of these areas to problems in AI and Statistics. He has co-authored the QSopt and QSopt_ex linear programming solvers and is an Area Editor of the Mathematical Programming Computation journal.

Event Time:
4:15pm – 5:15pm