Invited Talks

Tuesday 11 November, 10:30am, Forum 3

Louis-Martin Rousseau: Towards neuro-symbolic CP

This presentation explores the integration of machine learning into constraint programming (CP) to build the next generation of neuro-symbolic solvers. It highlights how learning-based techniques—such as reinforcement learning and graph neural networks—can enhance search heuristics and propagation strategies. The presentation also examines recent advances in leveraging Lagrangian relaxation and decomposition methods through learning to strengthen dual bounds and improve solver efficiency. Finally, it explores how AI can be used to enhance the art of modelling in CP, and it discusses the many challenges of combining artificial intelligence and logical reasoning to enhance the power, adaptability, and scalability of combinatorial optimization.

Louis-Martin Rousseau is a Professor in the Department of Mathematics and Industrial Engineering at Polytechnique Montréal. An internationally recognized expert in artificial intelligence, operations research, and management science, his work focuses on combinatorial optimization, column generation, transportation logistics, scheduling, and healthcare resource optimization. Since 2016, he has held the Canada Research Chair in Healthcare Logistics (HANALOG), where his research aims to improve the planning and efficiency of healthcare services through advanced optimization and AI-driven decision-support systems.

Wednesday 12 November, 9:00am (Forum 1, joint with ICAPS)

Hanna Kurniawati: Sequential Decision-Making for Robots Operating in Non Deterministic and Partially Observable World

In recent years, robotics hardware has advanced tremendously, with increasingly affordable humanoids, quadrupeds, telepresence robots, and many more. Despite these advances, developing autonomous or semi-autonomous robots that can reliably, efficiently, and safely operate in our environments remains an open problem. Key to this difficulty is the ubiquity of uncertainty. These robots must compute effective strategies to achieve their goals even when the outcomes of their actions are uncertain, their sensors and perception systems are erroneous, and the environments they operate in are dynamic and only partially observable. Moreover, they must ensure safety for both the robots and the humans around them. However, the technology that enables robots to efficiently construct effective strategies in the presence of a wide variety of uncertainty is still lacking. In this talk, I will present some of our work in developing such a technology, specifically on our recent work in Partially Observable Markov Decision Processes (POMDPs) —the general and principled framework for sequential decision-making under uncertainty. I will also present how this technology can be applied for safety assurance of autonomous systems.

Hanna Kurniawati is a Professor at the ANU School of Computing and holds the SmartSat Chair for System Autonomy, Intelligence & Decision-Making. Hanna’s research spans robotics, decision-making under uncertainty, motion planning, computational geometry applications, integrated planning and learning, and reinforcement learning. Her works on scalable methods for planning under uncertainty have received multiple recognitions, including a student best paper award at ICAPS’15, a finalist for the best paper award at ICRA’15, and the RSS’21Test of Time Award. She has given keynote talks at IROS’18 and ICRA’25. Hanna was a Senior Editor of IEEE RA-L, the Award Chair of CoRL’22, a Program Co-Chair of ICRA’22, and is an Editor of IEEE TRO.

Thursday 13 November, 9:00am (Forum 1, joint with ICAPS)

Bistra Dilkina: Machine Learning meets Combinatorial Optimization

Machine learning and discrete optimization have both made significant strides in methodology and in successful applications. Their fusion, however, can provide the next big step change in solving hard combinatorial optimization problems more effectively. In this talk, I will highlight recent successes in integrating ML into existing combinatorial optimization algorithms, using distributions of related optimization instances as training data and leveraging techniques such as contrastive loss and multi-task learning. The successful hybridization of neural models and symbolic solvers will be demonstrated across mixed integer linear programming, multi-agent path finding, and nonlinear optimization problems.

Bistra Dilkina is an Associate Professor of Computer Science and the Dr. Allen and Charlotte Ginsburg Early Career Chair in Computer Science at the University of Southern California, USA. Her research focuses on challenging computational problems in sustainability and sustainable development, particularly decision and optimization problems. She is interested in network design problems as they arise in large-scale wildlife conservation planning and urban planning. She is a USC CREATE research fellow and co-Director of the USC Center for AI in Society (CAIS), a joint effort between the USC Viterbi School of Engineering and the USC Suzanne Dworak-Peck School of Social Work. During 2013-2017, she was as an Assistant Professor in the College of Computing at the Georgia Institute of Technology and a co-director of the Data Science for Social Good Atlanta summer program.

Google Sites

Report abuse