ELLIS Delft Talk by Neil Yorke-Smith: Linear and Bi-Linear Mixed Integer Formulations of Graph Neural Networks
02 April 2024 16:00 till 17:00 - Location: Hybrid: Building 28, Room Hilbert (2.W510) / Zoom - By: ELLIS Delft | Add to my calendar
by Neil Yorke-Smith | Delft University of Technology
Abstract
ReLU neural networks have been modelled as constraints in mixed integer linear programming (MILP), enabling surrogate-based optimisation in various domains and efficient solution of machine learning certification problems. However, previous works are mostly limited to MLPs. Graph neural networks (GNNs) can learn from non-euclidean data structures such as molecular structures efficiently and are thus highly relevant to computer-aided molecular design (CAMD), for example. We propose a bilinear formulation for ReLU Graph Convolutional Neural Networks and a MILP formulation for ReLU GraphSAGE models. These novel formulations enable solving optimisation problems with embedded trained GNNs to global optimality. We apply our optimization approach to an illustrative CAMD case study where the formulations of the trained GNNs are used to design molecules with optimal boiling points. This is joint work with T. McDonald, C. Tsay and A.M. Schweidtmann.
Speaker Biography
Neil Yorke-Smith directs the Socio-Technical Algorithmic Research (STAR) Lab at TU Delft. His research addresses a fundamental question of the AI era: how can technology help people make decisions in complex socio-technical situations? Yorke-Smith is currently an Associate Professor in the Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS/EWI), Delft University of Technology. Previously, Yorke-Smith held positions at the American University of Beirut, Lebanon, and SRI International, USA. Yorke-Smith served as General Chair of BNAIC/BeNeLearn'23, Programme Chair of IAAI'21 and of AAMAS'20, Area Chair of ECAI'23, and Proceedings Chair of RecSys'23; and currently sits on the Editorial Boards of the journals AI, Constraints, JAAMAS and JAIR. Yorke-Smith is a Senior Member of AAAI, a Senior Member of ACM, and a member of CLAIRE and ELLIS. More information and publications: starlab.ewi.tudelft.nl.