{"id":"arxiv/2603.20891","url":"https://arxiv.org/abs/2603.20891","tags":["stat.ML","cs.LG","eess.SP","math.DS"],"score":0,"title":"Auto-differentiable data assimilation: Co-learning of states, dynamics, and filtering algorithms","domain":"stat.ML","source":"arxiv","authors":["Melissa Adrian","Daniel Sanz-Alonso","Rebecca Willett"],"reviews":0,"subject":"stat.ML","summary":"Data assimilation algorithms estimate the state of a dynamical system from partial observations, where the successful performance of these algorithms hinges on costly parameter tuning and on employing an accurate model for the dynamics. This paper introduces a framework for jointly learning the state, dynamics, and parameters of filtering algorithms in data assimilation through a process we refer to as auto-differentiable filtering. The framework leverages a theoretically motivated loss function that enables learning from partial, noisy observations via gradient-based optimization using…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-03-21","reactions":{}}