Combinatorial Optimization Enhances Trustworthy Machine Learning.

A new survey paper examines the intersection of combinatorial optimization (CO) and trustworthy machine learning (ML), proposing CO as a powerful framework for addressing critical ML properties. It argues that while modern ML models are complex, CO can help characterize their behavior beyond mere empirical performance, focusing on transparency, interpretability, robustness, fairness, privacy, and certifiability. The review synthesizes recent advancements where CO formulations offer distinct advantages over purely heuristic or gradient-based methods. These advantages include providing global guarantees, formal certificates, and explicit management of trade-offs across various tasks, from interpretable model learning and explanation generation to robustness analysis and privacy protection. Despite ongoing challenges related to scalability, the paper suggests that continuous progress in solvers and hybrid algorithms indicates a growing and significant role for combinatorial optimization in the design and deployment of reliable and trustworthy AI systems.