Research / Application AreasMachine Learning and AI for Science
Machine Learning and Artificial intelligence enable the learning of complex nonlinear patterns from high-dimensional datasets. In ESAM, we are interested in leveraging or developing new data-driven methods for scientific learning. Scientific data, especially from dynamical systems, is often smaller and structurally different from the data that many ML algorithms were developed to analyze. Additionally, science, engineering, and medicine demand uncertainty quantification and low prediction error. Many mathematical challenges arise when learning scientifically interpretable models, incorporating physical or biological principles as constraints during learning, and evaluating the robustness and reliability of the methods. We make connections between data-assimilation, sparse-nonlinear optimization, statistical physics, information theory, dynamical systems, manifold learning, and ML-based classification and prediction algorithms. Together we are pushing the boundaries of how to learn from scientific data to enhance our understanding of the physical, biological, and social world.