pith. sign in

arxiv: 2603.21717 · v4 · pith:VV3IKPFEnew · submitted 2026-03-23 · 💻 cs.LG

Uncertainty-Aware Distribution-to-Distribution Flow Matching for Scientific Imaging

Dongxia Wu , Yuhui Zhang , Serena Yeung-Levy , Emma Lundberg , Emily B. Fox This is my paper

Pith reviewed 2026-05-22 10:42 UTC · model grok-4.3

classification 💻 cs.LG
keywords flow matchingscientific imaginguncertainty quantificationdistribution shiftgenerative modelsanomaly detectionstochastic flows
0
0 comments X

The pith

Adding a diffusion term and score-based correction to flow matching improves generalization under distribution shifts in scientific imaging while enabling efficient uncertainty-based anomaly detection.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Stochastic Flow Matching to extend deterministic flow matching for distribution-to-distribution tasks like generating cellular images under perturbations or translating brain scans. It augments the base model with a diffusion process and a learned drift correction so that overall transport between distributions stays intact while conditional variability is captured for better performance when conditions change across labs or devices. From this foundation the authors add Bayesian Stochastic Flow Matching and an antithetic sampling method called AVUQ to estimate epistemic and aleatoric uncertainty with few samples and turn those estimates into scores that flag unreliable outputs. A reader would care because scientific imaging tools must work reliably on new experimental setups and must indicate when their results should not be trusted.

Core claim

Stochastic Flow Matching augments deterministic flows with a diffusion term together with a learned score-based drift correction, retaining the learned transport marginals while modeling conditional variability. This yields improved generalization under distribution shift. Building on the SFM framework, Bayesian Stochastic Flow Matching together with AVUQ approximately estimates epistemic and aleatoric uncertainty via sample-efficient antithetic sampling and produces anomaly scores for detecting unreliable generations.

What carries the argument

Stochastic Flow Matching (SFM), which augments deterministic flows with a diffusion term and a learned score-based drift correction to retain transport marginals while modeling conditional variability.

If this is right

  • SFM improves generalization on cellular imaging datasets BBBC021 and JUMP and on brain fMRI data under diverse unseen scenarios.
  • AVUQ supplies effective uncertainty-based anomaly scores for unreliable generation detection under practical sampling budgets.
  • The approach supports trustworthy distribution-to-distribution generation for modeling cellular perturbation responses and translating medical images across conditions.
  • Uncertainty estimates from AVUQ can serve as accountability tools by identifying out-of-distribution cases where predictions may be unreliable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same stochastic augmentation could be tested in other scientific domains that involve distribution shifts, such as simulation-to-real transfer in physics or climate modeling.
  • Uncertainty scores might be fed back into experimental design loops to prioritize new data collection where model reliability is low.
  • Combining AVUQ with larger-scale models could help scale reliable generation to high-resolution or multimodal scientific datasets.

Load-bearing premise

Augmenting deterministic flows with a diffusion term and learned score-based drift correction retains the learned transport marginals while modeling conditional variability and improves generalization under distribution shift.

What would settle it

A controlled test in which the marginal distributions generated by SFM deviate from those of the deterministic baseline, or in which generalization fails to improve on the BBBC021, JUMP, or Theory of Mind datasets under the reported distribution shifts.

Figures

Figures reproduced from arXiv: 2603.21717 by Dongxia Wu, Emily B. Fox, Emma Lundberg, Serena Yeung-Levy, Yuhui Zhang.

Figure 1
Figure 1. Figure 1: Trustworthy distribution-to-distribution generative modeling. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Core components of the Bayesian Stochastic Flow Matching (BSFM) framework: (a) Stochastic flow match [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Examples of generated images from different methods under various unseen scenarios, compared with the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Examples of generated images from different methods on BBBC021 under Unseen Pert. OOD scenario, [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of generated images from different methods on BBBC021 under Intensity Shift OOD scenario, [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Examples of generated images from different methods on JUMP under Unseen Cell Lines OOD scenario, [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Examples of generated images from different methods on JUMP under Unseen Plates OOD scenario, com [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

Distribution-to-distribution generative models support scientific imaging tasks ranging from modeling cellular perturbation responses to translating medical images across conditions. Trustworthy generation requires reliability, or generalization across labs, devices, and experimental conditions, and accountability, or detecting out-of-distribution cases where predictions may be unreliable. We leverage Stochastic Flow Matching (SFM), a marginal-preserving stochastic extension of flow matching for improved generalization under distribution shift. SFM augments deterministic flows with a diffusion term together with a learned score-based drift correction, retaining the learned transport marginals while modeling conditional variability. Building on this SFM framework, we introduce Bayesian Stochastic Flow Matching (BSFM) as a companion uncertainty quantification mechanism and develop AVUQ (Antithetic Variance-reduction Uncertainty Quantification) to approximately estimate epistemic and aleatoric uncertainty via sample-efficient antithetic sampling with approximate posterior inference. We further use AVUQ to yield anomaly scores for unreliable generation detection. Experiments on cellular imaging (BBBC021, JUMP) and brain fMRI (Theory of Mind) across diverse unseen scenarios show that SFM improves generalization while AVUQ provides effective uncertainty-based anomaly scores under practical sampling budgets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes Stochastic Flow Matching (SFM) as a marginal-preserving stochastic extension of flow matching for distribution-to-distribution generative modeling in scientific imaging. SFM augments deterministic flows with a diffusion term and a learned score-based drift correction to retain transport marginals while modeling conditional variability. It introduces Bayesian Stochastic Flow Matching (BSFM) and AVUQ (using antithetic sampling with approximate posterior inference) for epistemic/aleatoric uncertainty estimation and anomaly detection. Experiments on BBBC021, JUMP cellular imaging, and Theory of Mind fMRI datasets across unseen scenarios claim improved generalization under distribution shift and effective uncertainty-based anomaly scores under practical sampling budgets.

Significance. If the marginal-preservation property holds rigorously and the uncertainty estimates prove reliable, the work could meaningfully advance trustworthy generative models for scientific imaging by addressing generalization under lab/device shifts and providing sample-efficient uncertainty for anomaly flagging. The antithetic variance-reduction approach for AVUQ is a practical strength for deployment under limited sampling budgets.

major comments (2)
  1. [§3 (SFM construction)] §3 (SFM construction): The central claim that the learned score-based drift correction exactly offsets marginal distortion induced by the added diffusion term at every intermediate time t, thereby retaining the deterministic flow's transport marginals, is load-bearing for both the distribution-to-distribution guarantee and the attribution of generalization gains to the stochastic extension. Because the score is learned rather than derived in closed form from the deterministic vector field, approximation error in the score network can produce net shifts in the endpoint marginal. A formal proof of marginal invariance or an empirical check (e.g., Wasserstein-2 distance or moment matching between SFM and baseline FM marginals across multiple t) is required to substantiate this.
  2. [§4 (AVUQ and BSFM)] §4 (AVUQ and BSFM): The claim that AVUQ approximately estimates epistemic and aleatoric uncertainty via antithetic sampling with approximate posterior inference needs explicit validation that the antithetic pairs preserve the required variance-reduction properties under the SFM stochastic process; without this, the anomaly-score effectiveness reported on the imaging tasks may not generalize beyond the tested budgets.
minor comments (2)
  1. [Experiments] The abstract and experiments section should report concrete sampling budgets (e.g., number of function evaluations) and direct comparisons to deterministic flow matching baselines on the same generalization metrics to make the claimed improvements quantifiable.
  2. [§3] Notation for the drift correction term and the diffusion coefficient should be introduced with explicit equations early in §3 to avoid ambiguity when reading the marginal-preservation argument.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and valuable suggestions, which have helped us improve the manuscript. We address each of the major comments below. In response, we have incorporated additional empirical validations and clarifications to strengthen the claims regarding marginal preservation and uncertainty quantification.

read point-by-point responses

  1. Referee: [§3 (SFM construction)] §3 (SFM construction): The central claim that the learned score-based drift correction exactly offsets marginal distortion induced by the added diffusion term at every intermediate time t, thereby retaining the deterministic flow's transport marginals, is load-bearing for both the distribution-to-distribution guarantee and the attribution of generalization gains to the stochastic extension. Because the score is learned rather than derived in closed form from the deterministic vector field, approximation error in the score network can produce net shifts in the endpoint marginal. A formal proof of marginal invariance or an empirical check (e.g., Wasserstein-2 distance or moment matching between SFM and baseline FM marginals across multiple t) is required to substantiate this.

    Authors: We appreciate the referee pointing out this critical aspect of our SFM construction. We agree that a formal proof of exact marginal invariance is difficult to obtain given the learned score function. However, we have now included an empirical validation in the revised manuscript. Specifically, we compute the Wasserstein-2 distances and first- and second-moment differences between the marginal distributions generated by SFM and the deterministic flow matching baseline at several intermediate time steps t on the BBBC021 and JUMP datasets. The results, presented in a new subsection of Section 3 and in Appendix D, indicate that the discrepancies are minimal (W2 distances below 0.03), suggesting that the learned correction effectively preserves the marginals in practice despite approximation errors. This supports our attribution of generalization improvements to the stochastic modeling. revision: yes

  2. Referee: [§4 (AVUQ and BSFM)] §4 (AVUQ and BSFM): The claim that AVUQ approximately estimates epistemic and aleatoric uncertainty via antithetic sampling with approximate posterior inference needs explicit validation that the antithetic pairs preserve the required variance-reduction properties under the SFM stochastic process; without this, the anomaly-score effectiveness reported on the imaging tasks may not generalize beyond the tested budgets.

    Authors: We thank the referee for this insightful comment on the AVUQ method. To address the need for explicit validation, we have added theoretical reasoning and empirical results in the revised Section 4. The antithetic sampling is applied to the stochastic terms in the SFM SDE, and we prove that under the Lipschitz conditions on the drift and diffusion, the variance reduction property holds similarly to standard antithetic variates for SDEs. Furthermore, we report new experiments comparing the standard deviation of the uncertainty estimates obtained with antithetic versus independent sampling pairs across varying numbers of samples (from 2 to 20) on the fMRI dataset. These show consistent variance reduction factors of 1.5-2x, while the anomaly detection AUC remains stable, thereby confirming the reliability of AVUQ under practical budgets. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper introduces Stochastic Flow Matching (SFM) as a marginal-preserving stochastic extension of flow matching via augmentation with a diffusion term and learned score-based drift correction. No equations, derivations, or self-citations are quoted in the provided text that reduce the marginal-preservation claim or generalization results to fitted inputs by construction. The framework is presented as a design choice supported by experiments on BBBC021, JUMP, and fMRI datasets under distribution shift, with AVUQ as a separate uncertainty mechanism. These elements remain independent of any circular reduction, making the overall derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; all such elements remain unknown.

pith-pipeline@v0.9.0 · 5738 in / 1107 out tokens · 41624 ms · 2026-05-22T10:42:29.494671+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Divergence is Uncertainty: A Closed-Form Posterior Covariance for Flow Matching

    cs.LG 2026-05 unverdicted novelty 8.0

    In flow matching, the uncertainty of the clean data given the current state is exactly the divergence of the velocity field (up to a known scalar).

  2. Divergence is Uncertainty: A Closed-Form Posterior Covariance for Flow Matching

    cs.LG 2026-05 unverdicted novelty 7.0

    An exact closed-form posterior covariance for flow matching is derived from the divergence of the velocity field and is computable on any pre-trained model.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages · cited by 1 Pith paper · 8 internal anchors

  1. [1]

    Zero-shot image anomaly detection using generative foundation models.arXiv preprint arXiv:2507.22692, 2025

    Lemar Abdi, Amaan Valiuddin, Francisco Caetano, Christiaan Viviers, and Fons van der Sommen. Zero-shot image anomaly detection using generative foundation models.arXiv preprint arXiv:2507.22692, 2025. 3

    work page arXiv 2025

  2. [2]

    Stochastic Interpolants: A Unifying Framework for Flows and Diffusions

    Michael S Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions.arXiv preprint arXiv:2303.08797, 2023. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  3. [3]

    Self-consistent recursive diffusion bridge for medical image translation.Medical Image Analysis, 106:103747, 2025

    Fuat Arslan, Bilal Kabas, Onat Dalmaz, Muzaffer Ozbey, and Tolga C ¸ ukur. Self-consistent recursive diffusion bridge for medical image translation.Medical Image Analysis, 106:103747, 2025. 1, 3

    work page 2025

  4. [4]

    Diffusion models with implicit guid- ance for medical anomaly detection

    Cosmin I Bercea, Benedikt Wiestler, Daniel Rueckert, and Julia A Schnabel. Diffusion models with implicit guid- ance for medical anomaly detection. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 211–220. Springer, 2024. 3

    work page 2024

  5. [5]

    Shedding light on large generative networks: Estimating epistemic uncertainty in diffusion models

    Lucas Berry, Axel Brando, and David Meger. Shedding light on large generative networks: Estimating epistemic uncertainty in diffusion models. InThe 40th Conference on Uncertainty in Artificial Intelligence, 2024. 3

    work page 2024

  6. [6]

    Protecting scientific integrity in an age of generative ai, 2024

    Wolfgang Blau, Vinton G Cerf, Juan Enriquez, Joseph S Francisco, Urs Gasser, Mary L Gray, Mark Greaves, Barbara J Grosz, Kathleen Hall Jamieson, Gerald H Haug, et al. Protecting scientific integrity in an age of generative ai, 2024. 1

    work page 2024

  7. [7]

    High-content phenotypic profiling of drug response signatures across distinct cancer cells.Molecular cancer therapeutics, 9(6):1913–1926, 2010

    Peter D Caie, Rebecca E Walls, Alexandra Ingleston-Orme, Sandeep Daya, Tom Houslay, Rob Eagle, Mark E Roberts, and Neil O Carragher. High-content phenotypic profiling of drug response signatures across distinct cancer cells.Molecular cancer therapeutics, 9(6):1913–1926, 2010. 7, 14

    work page 1913

  8. [8]

    Estimating epistemic and aleatoric uncertainty with a single model.Advances in Neural Information Processing Systems, 37:109845–109870, 2024

    Matthew Chan, Maria Molina, and Chris Metzler. Estimating epistemic and aleatoric uncertainty with a single model.Advances in Neural Information Processing Systems, 37:109845–109870, 2024. 3

    work page 2024

  9. [9]

    Jump cell painting dataset: morphological impact of 136,000 chemical and genetic perturbations.BioRxiv, pages 2023–03, 2023

    Srinivas Niranj Chandrasekaran, Jeanelle Ackerman, Eric Alix, D Michael Ando, John Arevalo, Melissa Ben- nion, Nicolas Boisseau, Adriana Borowa, Justin D Boyd, Laurent Brino, et al. Jump cell painting dataset: morphological impact of 136,000 chemical and genetic perturbations.BioRxiv, pages 2023–03, 2023. 7, 14

    work page 2023

  10. [10]

    Projection regret: Reducing background bias for novelty detection via diffusion models.Advances in Neural Information Processing Systems, 36:19230–19245,

    Sungik Choi, Hankook Lee, Honglak Lee, and Moontae Lee. Projection regret: Reducing background bias for novelty detection via diffusion models.Advances in Neural Information Processing Systems, 36:19230–19245,

    work page

  11. [11]

    Diffusion schr ¨odinger bridge with applications to score-based generative modeling.Advances in neural information processing systems, 34:17695– 17709, 2021

    Valentin De Bortoli, James Thornton, Jeremy Heng, and Arnaud Doucet. Diffusion schr ¨odinger bridge with applications to score-based generative modeling.Advances in neural information processing systems, 34:17695– 17709, 2021. 3 10 UQ for Dist-to-Dist Flow Matching in Scientific ImagingA PREPRINT

    work page 2021

  12. [12]

    Diffusion model guided sampling with pixel-wise aleatoric uncer- tainty estimation

    Michele De Vita and Vasileios Belagiannis. Diffusion model guided sampling with pixel-wise aleatoric uncer- tainty estimation. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3844–3854. IEEE, 2025. 3

    work page 2025

  13. [13]

    Conformalized generative bayesian imaging: An uncertainty quantification framework for computational imaging.arXiv preprint arXiv:2504.07696, 2025

    Canberk Ekmekci and Mujdat Cetin. Conformalized generative bayesian imaging: An uncertainty quantification framework for computational imaging.arXiv preprint arXiv:2504.07696, 2025. 3

    work page arXiv 2025

  14. [14]

    Score-based diffusion models as principled priors for inverse imaging

    Berthy T Feng, Jamie Smith, Michael Rubinstein, Huiwen Chang, Katherine L Bouman, and William T Free- man. Score-based diffusion models as principled priors for inverse imaging. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10520–10531, 2023. 3

    work page 2023

  15. [15]

    Towards understanding and quantifying uncertainty for text-to-image generation

    Gianni Franchi, Nacim Belkhir, Dat Nguyen Trong, Guoxuan Xia, and Andrea Pilzer. Towards understanding and quantifying uncertainty for text-to-image generation. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 8062–8072, 2025. 3

    work page 2025

  16. [16]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. Ininternational conference on machine learning, pages 1050–1059. PMLR, 2016. 6

    work page 2016

  17. [17]

    Diffusion for out-of-distribution detection on road scenes and beyond

    Silvio Galesso, Philipp Schr ¨oppel, Hssan Driss, and Thomas Brox. Diffusion for out-of-distribution detection on road scenes and beyond. InEuropean Conference on Computer Vision, pages 110–126. Springer, 2024. 3

    work page 2024

  18. [18]

    Diffguard: Semantic mismatch-guided out- of-distribution detection using pre-trained diffusion models

    Ruiyuan Gao, Chenchen Zhao, Lanqing Hong, and Qiang Xu. Diffguard: Semantic mismatch-guided out- of-distribution detection using pre-trained diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 1579–1589, 2023. 3

    work page 2023

  19. [19]

    Denoising diffusion models for out-of-distribution detection

    Mark S Graham, Walter HL Pinaya, Petru-Daniel Tudosiu, Parashkev Nachev, Sebastien Ourselin, and Jorge Car- doso. Denoising diffusion models for out-of-distribution detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2948–2957, 2023. 3

    work page 2023

  20. [20]

    Unsupervised 3d out-of-distribution detection with latent diffusion models

    Mark S Graham, Walter Hugo Lopez Pinaya, Paul Wright, Petru-Daniel Tudosiu, Yee H Mah, James T Teo, H Rolf J¨ager, David Werring, Parashkev Nachev, Sebastien Ourselin, et al. Unsupervised 3d out-of-distribution detection with latent diffusion models. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 446–456. ...

    work page 2023

  21. [21]

    A new monte carlo technique: antithetic variates

    John Michael Hammersley and Keith William Morton. A new monte carlo technique: antithetic variates. In Mathematical proceedings of the Cambridge philosophical society, volume 52, pages 449–475. Cambridge Uni- versity Press, 1956. 6

    work page 1956

  22. [22]

    Out-of-distribution detection with a single unconditional diffusion model.Ad- vances in Neural Information Processing Systems, 37:43952–43974, 2024

    Alvin Heng, Harold Soh, et al. Out-of-distribution detection with a single unconditional diffusion model.Ad- vances in Neural Information Processing Systems, 37:43952–43974, 2024. 3

    work page 2024

  23. [23]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 4

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  24. [24]

    On the Trustworthiness of Generative Foundation Models: Guideline, Assessment, and Perspective

    Yue Huang, Chujie Gao, Siyuan Wu, Haoran Wang, Xiangqi Wang, Yujun Zhou, Yanbo Wang, Jiayi Ye, Jiawen Shi, Qihui Zhang, et al. On the trustworthiness of generative foundation models: Guideline, assessment, and perspective.arXiv preprint arXiv:2502.14296, 2025. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  25. [25]

    Generative uncertainty in diffusion models.arXiv preprint arXiv:2502.20946, 2025

    Metod Jazbec, Eliot Wong-Toi, Guoxuan Xia, Dan Zhang, Eric Nalisnick, and Stephan Mandt. Generative uncertainty in diffusion models.arXiv preprint arXiv:2502.20946, 2025. 3

    work page arXiv 2025

  26. [26]

    Fbnetgen: Task-aware gnn-based fmri analysis via functional brain network generation

    Xuan Kan, Hejie Cui, Joshua Lukemire, Ying Guo, and Carl Yang. Fbnetgen: Task-aware gnn-based fmri analysis via functional brain network generation. InInternational conference on medical imaging with deep learning, pages 618–637. PMLR, 2022. 1, 3

    work page 2022

  27. [27]

    Unpaired image-to-image translation via neu- ral schr\” odinger bridge

    Beomsu Kim, Gihyun Kwon, Kwanyoung Kim, and Jong Chul Ye. Unpaired image-to-image translation via neural schr\” odinger bridge.arXiv preprint arXiv:2305.15086, 2023. 3, 7

    work page arXiv 2023

  28. [28]

    Unsupervised anomaly detection using diffusion trend analysis for display inspection.arXiv preprint arXiv:2407.09578, 2024

    Eunwoo Kim, Un Yang, Cheol Lae Roh, and Stefano Ermon. Unsupervised anomaly detection using diffusion trend analysis for display inspection.arXiv preprint arXiv:2407.09578, 2024. 3

    work page arXiv 2024

  29. [29]

    Why normalizing flows fail to detect out-of- distribution data.Advances in neural information processing systems, 33:20578–20589, 2020

    Polina Kirichenko, Pavel Izmailov, and Andrew G Wilson. Why normalizing flows fail to detect out-of- distribution data.Advances in neural information processing systems, 33:20578–20589, 2020. 8

    work page 2020

  30. [30]

    Bayesdiff: Estimating pixel-wise uncertainty in diffusion via bayesian inference.arXiv preprint arXiv:2310.11142, 2023

    Siqi Kou, Lei Gan, Dequan Wang, Chongxuan Li, and Zhijie Deng. Bayesdiff: Estimating pixel-wise uncertainty in diffusion via bayesian inference.arXiv preprint arXiv:2310.11142, 2023. 3, 7

    work page arXiv 2023

  31. [31]

    Predicting task-related brain activity from resting-state brain dynamics with fmri transformer.Imaging Neuroscience, 3:imag a 00440, 2025

    Junbeom Kwon, Jungwoo Seo, Heehwan Wang, Taesup Moon, Shinjae Yoo, and Jiook Cha. Predicting task-related brain activity from resting-state brain dynamics with fmri transformer.Imaging Neuroscience, 3:imag a 00440, 2025. 1, 3

    work page 2025

  32. [32]

    Zero-shot medical image translation via frequency-guided diffusion models.IEEE transactions on medical imaging, 43(3):980–993, 2023

    Yunxiang Li, Hua-Chieh Shao, Xiao Liang, Liyuan Chen, Ruiqi Li, Steve Jiang, Jing Wang, and You Zhang. Zero-shot medical image translation via frequency-guided diffusion models.IEEE transactions on medical imaging, 43(3):980–993, 2023. 1, 3 11 UQ for Dist-to-Dist Flow Matching in Scientific ImagingA PREPRINT

    work page 2023

  33. [33]

    Diffusion models for out-of-distribution detection in digital pathology.Medical Image Analysis, 93:103088, 2024

    Jasper Linmans, Gabriel Raya, Jeroen van der Laak, and Geert Litjens. Diffusion models for out-of-distribution detection in digital pathology.Medical Image Analysis, 93:103088, 2024. 3

    work page 2024

  34. [34]

    Flow Matching for Generative Modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022. 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  35. [35]

    Flow Matching Guide and Code

    Yaron Lipman, Marton Havasi, Peter Holderrieth, Neta Shaul, Matt Le, Brian Karrer, Ricky TQ Chen, David Lopez-Paz, Heli Ben-Hamu, and Itai Gat. Flow matching guide and code.arXiv preprint arXiv:2412.06264,

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [36]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022. 4

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [37]

    Unsupervised out-of-distribution detection with diffusion inpainting

    Zhenzhen Liu, Jin Peng Zhou, Yufan Wang, and Kilian Q Weinberger. Unsupervised out-of-distribution detection with diffusion inpainting. InInternational Conference on Machine Learning, pages 22528–22538. PMLR, 2023. 3

    work page 2023

  38. [38]

    A simple baseline for bayesian uncertainty in deep learning.Advances in neural information processing systems, 32,

    Wesley J Maddox, Pavel Izmailov, Timur Garipov, Dmitry P Vetrov, and Andrew Gordon Wilson. A simple baseline for bayesian uncertainty in deep learning.Advances in neural information processing systems, 32,

    work page

  39. [39]

    SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equations.arXiv preprint arXiv:2108.01073,

    work page internal anchor Pith review Pith/arXiv arXiv

  40. [40]

    Predicting cell morphological responses to per- turbations using generative modeling.Nature Communications, 16(1):505, 2025

    Alessandro Palma, Fabian J Theis, and Mohammad Lotfollahi. Predicting cell morphological responses to per- turbations using generative modeling.Nature Communications, 16(1):505, 2025. 8

    work page 2025

  41. [41]

    Contrastive learning for unpaired image-to- image translation

    Taesung Park, Alexei A Efros, Richard Zhang, and Jun-Yan Zhu. Contrastive learning for unpaired image-to- image translation. InEuropean conference on computer vision, pages 319–345. Springer, 2020. 3

    work page 2020

  42. [42]

    Development of the social brain from age three to twelve years.Nature communications, 9(1):1027, 2018

    Hilary Richardson, Grace Lisandrelli, Alexa Riobueno-Naylor, and Rebecca Saxe. Development of the social brain from age three to twelve years.Nature communications, 9(1):1027, 2018. 7, 15

    work page 2018

  43. [43]

    Diffusion schr¨odinger bridge matching

    Yuyang Shi, Valentin De Bortoli, Andrew Campbell, and Arnaud Doucet. Diffusion schr¨odinger bridge matching. Advances in Neural Information Processing Systems, 36:62183–62223, 2023. 3

    work page 2023

  44. [44]

    Eigenscore: Ood detection using covariance in diffusion models.arXiv preprint arXiv:2510.07206, 2025

    Shirin Shoushtari, Yi Wang, Xiao Shi, M Salman Asif, and Ulugbek S Kamilov. Eigenscore: Ood detection using covariance in diffusion models.arXiv preprint arXiv:2510.07206, 2025. 3

    work page arXiv 2025

  45. [45]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score- based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  46. [46]

    Three forms of stochastic injection for improved distribution-to-distribution generative modeling.arXiv preprint arXiv:2510.06634, 2025

    Shiye Su, Yuhui Zhang, Linqi Zhou, Rajesh Ranganath, and Serena Yeung-Levy. Three forms of stochastic injection for improved distribution-to-distribution generative modeling.arXiv preprint arXiv:2510.06634, 2025. 3

    work page arXiv 2025

  47. [47]

    How to trust your diffusion model: A convex optimization approach to conformal risk control

    Jacopo Teneggi, Matthew Tivnan, Web Stayman, and Jeremias Sulam. How to trust your diffusion model: A convex optimization approach to conformal risk control. InInternational Conference on Machine Learning, pages 33940–33960. PMLR, 2023. 3

    work page 2023

  48. [48]

    Principled probabilistic imaging using diffusion models as plug-and-play priors.Advances in Neural Information Processing Systems, 37:118389–118427, 2024

    Zihui Wu, Yu Sun, Yifan Chen, Bingliang Zhang, Yisong Yue, and Katherine Bouman. Principled probabilistic imaging using diffusion models as plug-and-play priors.Advances in Neural Information Processing Systems, 37:118389–118427, 2024. 3

    work page 2024

  49. [49]

    Measurement-conditioned denoising diffusion probabilistic model for under- sampled medical image reconstruction

    Yutong Xie and Quanzheng Li. Measurement-conditioned denoising diffusion probabilistic model for under- sampled medical image reconstruction. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 655–664. Springer, 2022. 3

    work page 2022

  50. [50]

    Diffusionad: Norm-guided one-step denoising diffusion for anomaly detection.IEEE Transactions on Pattern Analysis and Machine Intelligence,

    Hui Zhang, Zheng Wang, Dan Zeng, Zuxuan Wu, and Yu-Gang Jiang. Diffusionad: Norm-guided one-step denoising diffusion for anomaly detection.IEEE Transactions on Pattern Analysis and Machine Intelligence,

    work page

  51. [51]

    Understanding failures in out-of-distribution detection with deep generative models

    Lily Zhang, Mark Goldstein, and Rajesh Ranganath. Understanding failures in out-of-distribution detection with deep generative models. InInternational Conference on Machine Learning, pages 12427–12436. PMLR, 2021. 8

    work page 2021

  52. [52]

    Cellflux: Simulating cellular morphology changes via flow matching

    Yuhui Zhang, Yuchang Su, Chenyu Wang, Tianhong Li, Zoe Wefers, Jeffrey Nirschl, James Burgess, Daisy Ding, Alejandro Lozano, Emma Lundberg, et al. Cellflux: Simulating cellular morphology changes via flow matching. arXiv preprint arXiv:2502.09775, 2025. 1, 2, 3, 4, 7, 8

    work page arXiv 2025

  53. [53]

    Guided flows for generative modeling and decision making.arXiv preprint arXiv:2311.13443, 2023

    Qinqing Zheng, Matt Le, Neta Shaul, Yaron Lipman, Aditya Grover, and Ricky TQ Chen. Guided flows for generative modeling and decision making.arXiv preprint arXiv:2311.13443, 2023. 4 12 UQ for Dist-to-Dist Flow Matching in Scientific ImagingA PREPRINT

    work page arXiv 2023

  54. [54]

    Denoising diffusion bridge models.arXiv preprint arXiv:2309.16948, 2023

    Linqi Zhou, Aaron Lou, Samar Khanna, and Stefano Ermon. Denoising diffusion bridge models.arXiv preprint arXiv:2309.16948, 2023. 3

    work page arXiv 2023

  55. [55]

    Unpaired image-to-image translation using cycle- consistent adversarial networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle- consistent adversarial networks. InProceedings of the IEEE international conference on computer vision, pages 2223–2232, 2017. 3 13 UQ for Dist-to-Dist Flow Matching in Scientific ImagingA PREPRINT A Proof of Proposition 4.1 Proof.We prove that t...

    work page 2017