Research map / Probabilistic and scientific modeling / Variational and generative sampling
Derivative-Informed Neural Operator Acceleration of Geometric MCMC for Infinite-Dimensional Bayesian Inverse Problems
Lianghao Cao, Thomas O'Leary-Roseberry, Omar Ghattas
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train the neural operator with Jacobian (derivative-informed) loss instead of conventional input-output loss | mMALA with neural operator surrogate · Diffusion-reaction and hyperelastic BIPs | effective sample generation speed: 18--40 times faster | Improved |
| use DINO-driven geometric MCMC instead of prior-geometry pCN | pCN · Diffusion-reaction and hyperelastic BIPs | effective sample generation speed: 60--97 times faster | Improved |
| use DINO-driven geometric MCMC instead of full geometric MCMC | mMALA · Diffusion-reaction and hyperelastic BIPs | effective sample generation speed: 3--9 times faster | Improved |
| add derivative-informed training | Reduced basis neural operator · Diffusion-reaction and hyperelastic BIPs | training sample generation cost for equal Jacobian accuracy: 16--25 times lower cost | Improved |
| add derivative-informed training | Reduced basis neural operator · Nonlinear diffusion-reaction | training sample cost to accelerate geometric MCMC: 166 times difference | Improved |
| use conventionally trained neural operator in mMALA (n_t=16000) instead of DINO (ablation) | NO-mMALA · Diffusion-reaction coefficient inversion | mean square jump (x1e-3): 11 -> 2.1 | Hurt |
| increase DINO training samples from 2000 to 16000 (ablation) | DINO-mMALA · Diffusion-reaction coefficient inversion | step size (x1e-2): 12 -> 15 | Improved |
| add delayed acceptance to the conventional neural operator mMALA (ablation) | NO-mMALA · Diffusion-reaction coefficient inversion | mean square jump (x1e-3): 2.1 -> 1.1 | Hurt |
| use geometric proposal (mMALA) instead of prior-geometry pCN | pCN · Diffusion-reaction coefficient inversion | mean square jump (x1e-3): 1 -> 12 | Improved |
Closest papers by meaning
- Light-Weight Diffusion Multiplier and Uncertainty Quantification for Fourier Neural Operators (NeurIPS 2025 · similarity 0.86)
- SBNO : Schrödinger Bridge Neural Operator for the Forward and Inverse Problems under Missing Information (NeurIPS 2026 · similarity 0.86)
- Adaptive Coordinate Transforms for Neural Operators (NeurIPS 2026 · similarity 0.85)
- Physics-Informed Deep Inverse Operator Networks for Solving PDE Inverse Problems (ICLR 2025 · similarity 0.85)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.85)
- Compute-optimal data scaling for neural surrogates via multi-fidelity training (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
