Many-Body Simulations and Machine Learning

Our research advances Many-Body Simulations and Machine Learning by integrating statistical-physical model systems, non-equilibrium many-body dynamics, and machine learning to uncover new pathways for intelligent computation and physical modeling. We study driven, out-of-equilibrium systems using Monte Carlo methods and many-body simulations, e.g. of the Ising model and active matter systems, to generate rich spatio-temporal data, which serve as the foundation for machine learning. Leveraging self-supervised and generative methods, we train models to predict complex dynamics, extract latent representations, and capture memory and causal structures inherent in non-equilibrium evolution.

A key focus is the interpretation of machine learning through physics, where concepts like phase transitions and criticality are shown to shape learned representations, bridging algorithmic behavior with physical principles. This synergy enables physics-based machine learning, where models are not only predictive but also interpretable and grounded in first principles. By combining machine learning with many-body simulations, we develop hybrid frameworks that enhance both simulation efficiency and scientific insight, advancing the understanding of emergent phenomena while enabling scalable, energy-efficient computing.