arXiv Machine Learning

A Dynamic Likelihood Approach to Filtering for Advection-Diffusion Dynamics

A Bayesian data assimilation scheme called the Dynamic Likelihood Filter (DLF) is extended to handle advection-diffusion problems via a split-step formulation. The method propagates probabilities along characteristics using observations and their statistics, allowing the posterior estimate to inherit phase information. In advection-dominated scenarios, the DLF outperforms other assimilation approaches, especially when observations are sparse and have low uncertainty, though its computational cost scales cubically with the number of observations over time.

arXiv Machine Learning
1d ago

Auto-differentiable data assimilation: Co-learning of states, dynamics, and filtering algorithms

The paper presents a framework called auto‑differentiable filtering that jointly learns the state, dynamics, and filtering algorithm parameters in data assimilation. It uses a loss function that allows gradient‑based optimization from partial, noisy observations via auto‑differentiation. Experiments demonstrate the method on diverse dynamical systems, including aerospace, atmospheric, biological, and geophysical models, and the authors offer guidelines for tailoring the framework to specific observation models, accuracy needs, and computational budgets.

By Melissa Adrian, Daniel Sanz-Alonso, Rebecca Willett
arXiv Machine Learning
Jun 9

ForcingDAS: Unified and Robust Data Assimilation via Diffusion Forcing

arXiv:2605. 14285v2 Announce Type: replace-cross Abstract: Data assimilation (DA) estimates the state of an evolving dynamical system from noisy, partial observations, and is widely used in scientific simulation as well as weather and climate science.

By Yixuan Jia, Siyi Chen, Yida Pan, Xiao Li, Lianghe Shi, Chanyong Jung, Haijie Yuan, Ismail Alkhouri, Yue Cynthia Wu, Saiprasad Ravishankar, Jeffrey A Fessler, Qing Qu
arXiv Machine Learning
Oct 5

A Unified Framework for Bayesian Data Assimilation with Generative Models and Observation Interpolants

A new framework for Bayesian data assimilation is presented that leverages pretrained stochastic interpolants, flow matching, and diffusion models as posterior samplers without retraining. By conditioning the interpolant path on observations, a shared likelihood‑score correction is applied to the drift or velocity, unifying stochastic and deterministic sampling methods. The approach yields exact posterior samples when the intermediate likelihood score is known, and practical approximations use a Gaussian surrogate with bias‑corrected mean and inflated covariance, enabling tractable high‑dimensional applications such as linear‑Gaussian dynamics, stochastic Navier‑Stokes, and urban airflow simulations with up to $O(10^4)$ degrees of freedom.

By Nikolaj T. M\"ucke, Benjamin Sanderse
arXiv Machine Learning
Nov 6

A Survey of Nonlinear Filtering Methods for Data Assimilation and Recent Advances in Generative AI-Enabled Score Filters

The paper surveys nonlinear filtering methods for data assimilation, highlighting challenges in chaotic, high‑dimensional systems. It introduces the Ensemble Score Filter (EnSF), which encodes filtering densities via a score function and uses a training‑free Monte Carlo estimator to avoid Gaussian assumptions, particle collapse, and kernel scalability issues. The survey reviews EnSF’s mathematical foundations and fourteen extensions, noting that most retain the core estimator while two adapt it for planetary‑scale deployment.

By Cath Pham, Kadesha Reynolds