As discussed above, ab initio AI promises to greatly improve theoretical and experimental physics in ways that we are only beginning to explore. In turn, the unique features of these physics applications offer compelling research opportunities in AI more broadly. For example, physical systems often feature exact or approximate symmetries. Exploiting these known invariances in data structures and algorithms is a key to developing efficient algorithms for physics applications, but is also likely to be transformative in broader contexts. Physics-informed architectures and hardware development promise advances in the speed of AI algorithms, and work in statistical physics is providing a theoretical foundation for understanding AI dynamics. The IAIFI is supporting these efforts that deeply entwine our ab initio AI research with our ab inito physics goals.
IAIFI AI Projects: Embedding Quantum Field Theory Principles into AI, AI that Respects Algebraic and Geometric Invariance, Fast AI on Specialized Hardware, AI at the Speed of Light, Speeding up Simulations with Curiosity-Driven AI, From Sequential AI to Parallel AI, Statistical Physics for Sparse AI Learning, Combining Symmetries and Flexibility with AI, AI for Data/Simulation Differences, and Statistical Physics for AI Generalization.