The OECD.AI Policy Navigator

This program is a powerful tool supporting the national AI strategy, already encompassing nine ambitious projects. It aims to tackle key and fundamental challenges in machine learning across three complementary pillars:
 

Trustworthy and Distributed AI
Through a multifaceted approach to trust, this axis integrates:

• The development of robustness foundations using statistical approaches.
• The incorporation of formal methods for the specification, learning, and safety validation of AI models
• The development of decentralized learning mechanisms that ensure security.
• The integration of causal models with machine learning to provide explainable, evidence-based capabilities.

Frugal and Embedded AI
This involves deepening the theoretical and algorithmic foundations of machine learning to ensure data and computation frugality by design. It includes:

• Advanced model optimization and training guided by hardware constraints.
• The design of dedicated AI hardware architectures that are modular, flexible, and adaptive.

The exploration of physics-informed computational models.
New Mathematical Foundations of AI, particularly in the field of mathematical analysis, this pillar includes:

• The integration of tools and advances from partial differential equations, optimal control, and optimal transport.
• To develop novel architectures with stable optimization schemes, efficient solvers and approximations.
• The analytical treatment of generative and diffusion methods.