Equation Learner Network

Equation learner networks aim to automatically discover the underlying mathematical equations governing a system from observational data, a crucial task across numerous scientific disciplines. Current research focuses on improving the accuracy and efficiency of these networks, exploring techniques like sparse regression, Bayesian model selection, and the use of weak formulations to enhance robustness to noise and achieve better generalization. These advancements are impacting fields such as robotics (through improved skill extrapolation and adaptation), fluid dynamics (via subgrid-scale closure modeling), and potentially many others where uncovering governing equations from data is a limiting factor. The development of convexified learning approaches also promises to improve the reliability and interpretability of discovered equations.