Special sessions
In addition to submissions about general LION themes, we also welcome submissions related to one of our special sessions. The special sessions will be part of the regular conference and are subject to the same peer-review as all other submissions.
Special session 1: (Deep) Reinforcement Learning in OR Optimization
Organizer: Lin Xie1,2, Jinkyoo Park3, and Yaoxin Wu4
1Chair of Information Systems and Business Analytics, Brandenburg University of Technology, Germany
2University of Twente, Netherlands
3Korea Advanced Institute of Science and Technology (KAIST), Korea
4Eindhoven University of Technology, Netherlands
Abstract: The integration of Artificial Intelligence (AI) with Operations Research (OR) is transforming the way complex, large-scale, real-world optimization problems are solved across various industries. This special session will focus on the application of AI techniques, such as reinforcement learning and deep learning, to classical OR challenges in diverse fields, including logistics, transportation, robotics, manufacturing, energy, and healthcare. Topics include AI in combinatorial optimization and multi-agent optimization.
Special session 2: Enhancing Exact Combinatorial Optimization Solvers with Machine Learning
Organizer: Quentin Cappart1
1Polytechnique Montréal, Canada
Abstract: Combinatorial optimization finds applications in a wide range of fields, including aerospace, transportation planning, scheduling, and economics. In recent years, there has been growing interest in leveraging machine learning to solve combinatorial problems, either in an end-to-end manner or as a means to enhance existing exact solvers (e.g., integer programming, constraint programming, or SAT solvers).
The advantage of integrating machine learning with exact solvers lies in combining the solver's guarantees, such as proving optimality, with the flexibility of learning-based approaches. However, identifying the best methods to incorporate a learning module within a solver remains an open research question. This session focuses on papers that explore machine learning techniques to enhance combinatorial optimization solvers, such as learning how to branch, configuring solvers, pruning the search space, or related innovations.
Special session 3: Sustainability in surrogate models, Bayesian optimization, and parameter tuning
Organizer: Antonio Candelieri1 Laurens Bliek 2
1University of Milano-Bicocca, Italy
2Eindhoven University of Technology, Netherlands
Abstract: Nowadays, Artificial Intelligence is known for its potential to solve sustainability-related problems, as well as its negative impact on the climate due to its high energy consumption. At the same time, Bayesian optimization and other surrogate-based algorithms are receiving significant attention for their capability of efficiently solving optimization problems with expensive black-box objectives. This is common in parameter tuning problems in simulation optimization, automated machine learning, and automated algorithm configuration. By replacing the expensive objectives with a surrogate model, computational resources can be saved that would otherwise be allocated to running heavy physics simulators, to train deep learning models, or to run optimization algorithms on multiple problem instances. Therefore, this approach can save time and energy, and reduce emissions. Several challenges remain to properly take sustainability aspects into account with these techniques, such as sustainability measures, energy-aware search strategies, efficient surrogate models, transfer learning, and multi-objective/multi-fidelity optimization approaches. This session focuses on new developments to tackle these challenges, and also on new sustainability-related applications where surrogate-based algorithms such as Bayesian optimization are successfully applied.
Special session 4: Learning and Intelligent Optimization for Physical Systems
Organizer: Konstantinos Chatzilygeroudis 1,Michael Vrahatis 1
1University of Patras, Greece
Abstract: Working with physical systems presents unique challenges compared to theoretical models or simulations. Experimentation time must be minimized to avoid costly hardware failures, and algorithms must ensure the safety of both the system and surrounding humans. While data-driven approaches, such as Machine Learning (ML), can learn complex models and improve over time, they require large amounts of data and struggle to provide formal guarantees. Conversely, traditional optimization methods offer stronger theoretical guarantees with less data but lack adaptability for improving performance over time. This special session seeks submissions on "Learning and Intelligent Optimization for Physical Systems," focusing on methods that combine ML and optimization techniques to address real-world challenges. Topics include robot learning, embedded systems, real-time applications, and human-computer interaction.