Current research at the intersection of Combinatorial Optimization (CO) and Machine Learning (ML) is emerging as a game-changing force in Artificial Intelligence. This fusion of two traditionally distinct solving paradigms has recently gained strong momentum in the CO community and is becoming a ubiquitous technology across a wide range of applications. One area that can particularly benefit from ML is constraint reasoning and optimization.
Constraint Programming (CP) is widely regarded as one of the foremost paradigms for solving combinatorial problems in AI. CP provides a generic and expressive framework for modeling and solving problems arising in diverse application domains. However, the hybridization of Data Mining, Machine Learning, and Deep Learning with state-of-the-art discrete reasoning and optimization techniques remains one of the major challenges in AI.
This proposal addresses the challenge of data-driven decision making through a unified framework combining machine learning, discrete reasoning and optimization, and an associated solver based on Graphical Models (GMs), namely toulbar2—winner of the UAI 2022 solver competition on two discrete optimization and counting tasks. Discrete graphical models, particularly additive GMs such as Cost Function Networks (CFNs) and Markov Random Fields (MRFs), are especially attractive as they naturally and seamlessly integrate logical and probabilistic information. The toulbar2 solver is a state-of-the-art tool for reasoning over such models, supporting both exact algorithms and advanced meta-heuristics.
The GMLaS project represents a significant step toward the ambitious goal of integrating ML/DM tools within the Cost Function Network paradigm, leading to a hybrid, data-driven solving framework that natively addresses optimization problems and aims to scale to larger and more complex real-world applications. One key application targeted by the project is computational protein design (CPD), a domain in which one of the project partners has deep and well-established expertise.