Robust ensemble guidance improves scientific inverse problem reconstructions
Robust Ensemble Guidance for Scientific Inverse Problems
Machine Learning
Summary
Sometimes scientists try to figure out causes from complex effects using computer models, but the math can be tricky and easily thrown off by unusual data points. The authors show that by balancing how different measurements are weighted and cutting down the influence of extreme errors, their method makes these reverse calculations more reliable. This approach works without needing to change the original simulation tools and improves image and data reconstructions in tests like studying black holes and fluid flows. Their method helps handle tricky measurements better, leading to clearer and more accurate results.
What this means in practice
- •For scientific computing teams: Improve accuracy of image or data reconstructions when solving complex inverse problems involving simulations without altering simulation code.
- •For geophysics engineers: Enhance seismic data inversion processes to better estimate underground structures by controlling influence of measurement errors in ensemble approaches.
Authors
Zixiang Li, Wei Wang, Yunchao Wei, Yao Zhao, Yue Song
Abstract
Ensemble guidance combines pretrained diffusion priors with black-box forward models to solve inverse problems without differentiating through the physical simulator. However, observation coordinates with large predictive spread or extreme residuals can dominate the ensemble correction, degrading reconstruction accuracy. We show that two simple modifications, weighting and clipping, substantially improve this correction. Our method, Robust Ensemble Guidance (REG), uses ensemble predictive spread to balance observation scales and adaptively clips standardized residuals to limit the influence of extreme discrepancies. Both operations reuse existing particles and forward predictions, requiring no additional denoiser or forward-model evaluations. Under a local linear Gaussian model, we derive conditions for reduced one-step estimation risk, bound the influence of individual observation coordinates, and characterize when these benefits persist with finite ensembles. Experiments on Navier-Stokes inversion, black-hole imaging, and acoustic full-waveform inversion demonstrate improved reconstruction over the underlying ensemble solver. In particular, REG increases black-hole reconstruction PSNR by 6.2-8.2 dB across three observation regimes and reduces Navier-Stokes reconstruction error by 26.4\% in a matched-budget comparison. These findings highlight the importance of observation heterogeneity and residual influence in designing reliable generative solvers for scientific inverse problems.