Learning and Optimization

Algorithms and computation touch on all theoretical and practical aspects of computer science, both software and hardware. For the past decade, artificial intelligence and learning have focused on the automatic design of algorithms and computational processes, guided by data, experts, users, and/or the environment.

Research Topics

The A&O team—a joint Paris-Saclay, CNRS, and Inria Saclay project team—is interested in learning models from data, focusing on four fundamental areas.

• The first concerns adversarial learning, which is based on the interaction of two or more learning agents, replacing the unknown objective function with a min-max approach (game theory); this area is also crucial for the validation and certification of neural networks.
• The second area concerns the selection and configuration of a priori algorithms based on available data, also known as AutoML. This is not only a necessary condition for the democratization of AIArtificial Intelligence, but also a challenge that has remained unresolved for 40 years, linked to the definition of data order parameters.
• The third addresses the problems of learning complex models and deals with the identification of regularities that enable the well-founded augmentation of data in the many areas of application where data is small (small data) relative to complexity.
• Finally, far from replacing knowledge with models derived purely from data, one objective is to engage in dialogue with domain knowledge, expressed for example by partial differential equations. The challenge here is to bridge the gap between machine learning and numerical engineering, in collaboration with the Fluid Mechanics and Energy Department.