mlsanityData to 14 Oct 2026

Calibrated Test-Time Guidance for Bayesian Inference

arXiv 2602.22428 ↗First posted 25 Feb 2026cs.LG cs.AIcited by 5

On the map: Diffusion models → Inverse problems and image restoration

Abstract (arXiv)

Test-time guidance is a widely used mechanism for steering pretrained diffusion models toward outcomes specified by a reward function. Existing approaches, however, focus on maximizing reward rather than sampling from the true Bayesian posterior, leading to miscalibrated inference. In this work, we show that common test-time guidance methods do not recover the correct posterior distribution and identify the structural approximations responsible for this failure. We then propose consistent alternative estimators that enable calibrated sampling from the Bayesian posterior. We significantly outperform previous methods on a set of Bayesian inference tasks, and set a new state-of-the-art PSNR in black hole image reconstruction.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Placed in Inverse problems and image restoration, one of 14 directions in Diffusion models. Builds on 54 of its 82 references that are in the map. First posted on arXiv 25 Feb 2026.

Builds on

54 of 82 references.

Closest ancestors · 12

Same direction, newest first

  1. 2025InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences14 Mar
  2. 2024Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing1 Jul
  3. 2024Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors29 May
  4. 2024Fast Samplers for Inverse Problems in Iterative Refinement Models27 May
  5. 2024Divide-and-Conquer Posterior Sampling for Denoising Diffusion Priors18 Mar
  6. 2023Denoising Diffusion Models for Plug-and-Play Image Restoration15 May
  7. 2023A Variational Perspective on Solving Inverse Problems with Diffusion Models7 May
  8. 2022Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model1 Dec
  9. 2022Diffusion Posterior Sampling for General Noisy Inverse Problems29 Sep
  10. 2022Improving Diffusion Models for Inverse Problems using Manifold Constraints2 Jun
  11. 2022Denoising Diffusion Restoration Models27 Jan
  12. 2021Come-Closer-Diffuse-Faster: Accelerating Conditional Diffusion Models for Inverse Problems through Stochastic Contraction9 Dec

Other references · 42

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  3. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  4. 2021Diffusion Models Beat GANs on Image Synthesiscited by 8,255
  5. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  6. 2022Classifier-Free Diffusion Guidancecited by 5,847
  7. 2022Flow Matching for Generative Modelingcited by 5,702
  8. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,101
  9. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  10. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 3,045
  11. 2022Video Diffusion Modelscited by 2,305
  12. 2023Consistency Modelscited by 1,822
30 more
  1. 2022Building Normalizing Flows with Stochastic Interpolantscited by 995
  2. 2023Training Diffusion Models with Reinforcement Learningcited by 982
  3. 2022Equivariant Diffusion for Molecule Generation in 3Dcited by 677
  4. 2025Mean Flows for One-step Generative Modelingcited by 561
  5. 2023Universal Guidance for Diffusion Modelscited by 437
  6. 2024One Step Diffusion via Shortcut Modelscited by 391
  7. 2022DiffDock: Diffusion Steps, Twists, and Turns for Molecular Dockingcited by 313
  8. 2023FreeDoM: Training-Free Energy-Guided Conditional Diffusion Modelcited by 294
  9. 2022Diffusion Probabilistic Modeling for Video Generationcited by 238
  10. 2018Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flowscited by 221
  11. 2019Automatic Posterior Transformation for Likelihood-Free Inferencecited by 219
  12. 2021Benchmarking Simulation-Based Inferencecited by 162
  13. 2023Manifold Preserving Guided Diffusioncited by 157
  14. 2024Fine-Tuning of Continuous-Time Diffusion Models as Entropy-Regularized Controlcited by 143
  15. 2024Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-Based Decodingcited by 136
  16. 2024TFG: Unified Training-Free Guidance for Diffusion Modelscited by 107
  17. 2024Symbolic Music Generation with Non-Differentiable Rule Guided Diffusioncited by 54
  18. 2026One-step Latent-free Image Generation with Pixel Mean Flowscited by 53
  19. 2024Heavy-Tailed Diffusion Modelscited by 40
  20. 2026Meta Flow Maps enable scalable reward alignmentcited by 37
  21. 2025Skywork-VL Reward: An Effective Reward Model for Multimodal Understanding and Reasoningcited by 32
  22. 2016The Generalized Reparameterization Gradientcited by 30
  23. 2017Reducing Reparameterization Gradient Variancecited by 27
  24. 2023Free-form Flows: Make Any Architecture a Normalizing Flowcited by 26
  25. 2025Generative Uncertainty in Diffusion Modelscited by 23
  26. 2025Variational Control for Guidance in Diffusion Modelscited by 22
  27. 2014Variational Temperingcited by 20
  28. 2025Test-time scaling of diffusions with flow mapscited by 16
  29. 2019Likelihood-free MCMC with Amortized Approximate Ratio Estimatorscited by 10
  30. 2023Learning Distributions on Manifolds with Free-Form Flowscited by 7

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026Hierarchical Variational Policies for Reward-Guided Diffusioncited by 0
  2. 2026Variational Test-time Optimization for Diffusion Synchronizationcited by 0
  3. 2026Accelerated Likelihood Maximization for Diffusion-based Versatile Content Generationcited by 0
  4. 2026Training-Free Inpainting Across Domains with a Frozen Text-to-Image Diffusion Modelcited by 0
  5. 2026Characterizing and Mitigating Distributional Bias in Physics-Informed Generative Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Optimizing Diffusion Priors in Image Reconstruction from a Single Observation
  2. 2026A Stabilized Path-Space Approach to Diffusion-Based Posterior Sampling
  3. 2026Measuring Effective Data Resolution in Guided Diffusion Posteriors
  4. 2026Tempered Guided Diffusion
  5. 2026Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models
  6. 2023Variational Bayesian Imaging with an Efficient Surrogate Score-based Prior
  7. 2026Sample-efficient evidence estimation of score based priors for model selection
  8. 2024Bayesian Conditioned Diffusion Models for Inverse Problems