Presents a likelihood-free transport map learned by minimizing an averaged energy-distance objective to amortize Bayesian inference for inverse problems, including PDE-constrained cases with neural operator representations.
Title resolution pending
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
years
2026 4verdicts
UNVERDICTED 4representative citing papers
Nudging with learned surrogate dynamics converges exponentially to an explicit error floor determined by surrogate error and observation noise, with training data requirements quantified for noise-free cases.
A Bayesian filtering extension of statFEM assimilates sequential observational data into elastodynamic finite element models by modeling uncertainties as Gaussian random fields, advancing the state with a stochastic Newmark scheme, and approximating the non-Gaussian prior via perturbation to obtain,
FLUID uses a recurrent encoder to create a fixed-size summary of observations, then learns coupled forward and backward flows to approximate filtering distributions and recover smoothing paths for nonlinear dynamics, with support for extrapolation.
citing papers explorer
-
Amortized Energy-Based Bayesian Inference
Presents a likelihood-free transport map learned by minimizing an averaged energy-distance objective to amortize Bayesian inference for inverse problems, including PDE-constrained cases with neural operator representations.
-
Continuous Data Assimilation with Learned Surrogate Dynamics
Nudging with learned surrogate dynamics converges exponentially to an explicit error floor determined by surrogate error and observation noise, with training data requirements quantified for noise-free cases.
-
Statistical finite elements for sequential data synthesis in solid dynamics
A Bayesian filtering extension of statFEM assimilates sequential observational data into elastodynamic finite element models by modeling uncertainties as Gaussian random fields, advancing the state with a stochastic Newmark scheme, and approximating the non-Gaussian prior via perturbation to obtain,
-
FLUID: Flow-based Unified Inference for Dynamics
FLUID uses a recurrent encoder to create a fixed-size summary of observations, then learns coupled forward and backward flows to approximate filtering distributions and recover smoothing paths for nonlinear dynamics, with support for extrapolation.