Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Calibrated Test-Time Guidance for Bayesian Inference
Daniel Geyfman, Felix Draxler, Jan Niklas Groeneveld, Hyunsoo Lee, Theofanis Karaletsos, Stephan Mandt
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 |
|---|---|---|---|
| use CBG gradient-free estimator instead of posterior-mean/Gaussian guidance baselines | diffusion test-time guidance · Bayesian inference benchmark (avg of 5 tasks) | C2ST: 0.803 (DPS) -> 0.527 | Improved |
| increase compute (likelihood evaluations / samples K) for CBG (ablation) | CBG · Bayesian inference benchmark | C2ST | Improved |
| increase compute for baseline guidance methods (DPS, LGD, DPG, SCG) (ablation) | baseline guidance · Bayesian inference benchmark | C2ST | No clear effect |
| use gradient-free instead of gradient-based CBG estimator (ablation) | CBG · Bayesian inference benchmark (avg of 5 tasks) | C2ST: 0.681 -> 0.527 | Improved |
| increase number of samples K per step from 128 to 65536 (renoise) (ablation) | CBG gradient-free · Black hole imaging | PSNR: 25.36 -> 27.08 | Improved |
| replace DDPM inner loop with renoise one-step sampler (ablation) | CBG gradient-free, K=512 · Black hole imaging | PSNR: 26.10 -> 26.10; time ~23 h -> 2.9 min | No clear effect |
| CBG with large K versus test-time guidance baselines | CBG renoise K=65536 · Black hole imaging | PSNR: 26.07 (PnP-DM) -> 27.08 | Improved |
| CBG vs likelihood-free methods | CBG gradient-free · Bayesian inference benchmark | C2ST: 0.545 (SNPE) -> 0.527 | Improved |
Closest papers by meaning
- Tempered Guided Diffusion (NeurIPS 2026 · similarity 0.86)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.86)
- Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors (NeurIPS 2024 · similarity 0.85)
- Feedback Guidance of Diffusion Models (NeurIPS 2025 · similarity 0.85)
- Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models (NeurIPS 2024 · similarity 0.85)
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · 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.
