Séminaire

Date : 25 novembre 2025 14:00 - Salle :Salle A102 An Introduction to DC Programming and DCA: A unifying nonconvex optimization framework for AI and Beyond Le Thi Hoai An - Université de Lorraine

Difference of Convex (DC) programming and the DC Algorithm (DCA) provide a general and unifying framework for addressing nonconvex and nonsmooth optimization problems. Celebrating 40 years of development, DC programming and DCA have evolved into mature, powerful, and versatile tools with strong theoretical foundations and wide practical impact. This introductory talk presents the fundamental ideas, motivations, and key properties of DC programming and DCA, aimed at a broad computer science audience.
We review the major milestones in the evolution of the methodology and introduce a family of DCA-based algorithms designed for diverse applications, including logistics, communication
systems, finance, healthcare, cryptology, image processing, and robotics, … These examples illustrate how DCA offers scalable, robust, and efficient solutions to challenging real-world
nonconvex problems.

The talk concludes by showing how DC programming and DCA naturally integrate with modern AI methodologies. We discuss how DCA enhances the scalability of large learning models,
strengthens robustness in decision-making under uncertainty, and provides efficient solvers for nonconvex machine learning formulations. Several recent AI applications highlight the potential of combining DCA with generative models, reinforcement learning, and data-driven optimization, demonstrating the broad impact of DC programming and DCA in AI and beyond