Speaker: Dr Fulya Trösser is a Mathematical Optimisation Engineer at N-SIDE

Title: Exact Methods For Bayesian Network Structure Learning

Synopsis: Discrete Graphical Models (GMs) represent joint functions over large sets of discrete variables as a combination of smaller functions. There exist several instantiations of GMs, such as directed probabilistic GMs like Bayesian Networks (BNs), which have found a wide range of applications in fields such as bioinformatics, image processing, and risk analysis. In this work, we make contributions to the state of the art in learning the structure of BNs, namely the Bayesian Network Structure Learning problem (BNSL). We discover a new point in the design space of search algorithms, which achieves a different trade-off between inference strength and speed of inference. Existing algorithms for it opt for either maximal strength of inference, like the algorithms based on Integer Programming (IP) and branch-and-cut, or maximal speed of inference, like the algorithms based on Constraint Programming (CP). We specify properties of a specific class of inequalities, called cluster inequalities, which lead to an algorithm that performs much stronger inference than that based on CP, much faster than that based on IP. We combine this with novel ideas for stronger propagation and more compact domain representations to achieve state-of-the-art performance in the open-source solver ELSA (Exact Learning of bayesian network Structure using Acyclicity reasoning).

Bio: Dr Fulya Trösser is a Mathematical Optimisation Engineer at N-SIDE, a leading provider of innovative analytics solutions to organisations in the fields of life sciences and energy. She holds a PhD in Applied Mathematics and Computer Science from the University of Toulouse, where she developed exact methods for the Bayesian network structure learning problem. Her research interests lie in the potential combination of different approaches, having focused on Integer Programming in her BSc and MSc, and on Constraint Programming in her PhD. She has a great passion for the application of mathematical tools to real-life problems in areas of benefit to society.

WED 15 JUNE 4PM – 5PM AEST MELBOURNE | 8AM – 9AM CEST BRUSSELS

ZOOM MEETING ID: 873 1557 5255; PASSWORD: 778635