FreeBridge: Variational Schr\"odinger Bridges for Cellular Transition Dynamics

Chenyu You; Haibin Ling; Jun Ma; Qin Ren; Xurui Wang

REVIEW 2 major objections 1 minor 25 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

FreeBridge casts single-cell perturbation modeling as a Schrödinger bridge on a fixed manifold of instance-segmented cell shapes to keep intermediate states inside observed morphologies.

2026-06-27 14:00 UTC pith:N23M4ZUW

load-bearing objection FreeBridge adds a segmented-cell manifold and latent support regularization to Schrödinger Bridges for endpoint-only single-cell trajectories, but the abstract supplies no numbers or ablations so the biological payoff remains unverified. the 2 major comments →

arxiv 2606.11286 v1 pith:N23M4ZUW submitted 2026-06-09 cs.LG cs.AI

FreeBridge: Variational Schr\"odinger Bridges for Cellular Transition Dynamics

Xurui Wang , Qin Ren , Jun Ma , Haibin Ling , Chenyu You This is my paper

classification cs.LG cs.AI

keywords single-cell imagingSchrödinger bridgesperturbation modelinggenerative modelscellular dynamicshigh-content imagingvariational inference

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces FreeBridge to infer continuous cellular trajectories under chemical or genetic perturbations when only separate control and treated populations are observed in fixed images. It formulates the problem as a variational Schrödinger bridge whose states are the instance-segmented single-cell representations, which define a fixed cellular manifold, and adds empirical latent support regularization so that the learned stochastic transport stays within the support of real cell morphologies. Endpoint matching alone permits many paths that cross unsupported regions; the regularization is meant to rule those out while still reaching the observed treated population. The method is tested on three high-content imaging datasets under a unified protocol and reports maintained or improved endpoint fidelity together with mechanism-of-action retention and, on one dataset, fewer intermediate support violations.

Core claim

FreeBridge defines atomic states as instance-segmented single-cell representations, establishing a fixed cellular manifold, and learns stochastic transport constrained within this geometry via empirical latent support regularization, achieving competitive endpoint fidelity and mechanism-of-action retention while reducing intermediate support violations on BBBC021.

What carries the argument

Variational Schrödinger bridge with empirical latent support regularization on an instance-segmented single-cell manifold

Load-bearing premise

Defining atomic states via instance-segmented single-cell representations creates a fixed cellular manifold whose empirical latent support regularization is sufficient to guarantee biologically interpretable intermediate evolution.

What would settle it

Generated intermediate cell states that systematically lack close matches among the observed single-cell morphologies in the same datasets would show that the support regularization failed to enforce the claimed geometric constraint.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

The approach yields competitive or improved endpoint fidelity and mechanism-of-action retention on BBBC021, RxRx1, and JUMP under a single evaluation protocol.
On BBBC021 it produces fewer intermediate support violations than prior generative models that match only the endpoint marginals.
Geometric grounding through the fixed manifold is presented as necessary for obtaining biologically interpretable perturbation dynamics.

Where Pith is reading between the lines

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

The same manifold-plus-regularization construction could be applied to other trajectory-inference tasks where only marginal distributions are available and the observations live in a high-dimensional shape space.
If the manifold construction proves robust, it could serve as a modular component inside larger models that combine imaging with additional modalities such as gene expression.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

FreeBridge adds a segmented-cell manifold and latent support regularization to Schrödinger Bridges for endpoint-only single-cell trajectories, but the abstract supplies no numbers or ablations so the biological payoff remains unverified.

read the letter

The one or two things to know are that FreeBridge adds a manifold from segmented single cells and a latent support regularization to Schrödinger Bridges to model transitions from endpoint data only, and it claims to match or beat baselines on endpoint fidelity while reducing support violations on BBBC021.

The new part is the specific way they ground the transport in the cellular manifold using instance segmentation and then regularize the latent support to avoid out-of-manifold paths. This addresses a real issue in these models where paths can go through unsupported regions. Applying it across three datasets with a consistent protocol is a solid move.

It does well at identifying the problem of boundary consistency not determining intermediates and proposing a geometric fix. The project page might have more, but based on the abstract the idea is clearly motivated by the biology of fixed cells.

The soft spots are bigger than minor. Only the abstract is here, so no quantitative results, no ablations showing the regularization's contribution, and no validation that lower support violations translate to better biological trajectories. The weakest assumption is that the empirical support on the segmented manifold is enough for interpretable evolution; without checks against independent markers, the improvement could be superficial. The stress-test concern holds up because the paper equates the geometric metric with biological fidelity without extra evidence.

This is for computational biologists working on perturbation modeling in imaging assays. A reader who needs a method to generate plausible cell trajectories from marginals might find the formulation useful, but it needs the full details to assess.

I would recommend sending it for peer review. The core formulation is worth a look from referees who know both transport models and single-cell data, even if revisions will be needed for the experiments.

Referee Report

2 major / 1 minor

Summary. The paper proposes FreeBridge, a variational Schrödinger Bridge method for inferring stochastic cellular transition dynamics from high-content imaging perturbation assays observed only as separate marginals. Atomic states are defined via instance-segmented single-cell representations to establish a fixed cellular manifold, with empirical latent support regularization used to constrain transport within this geometry. On BBBC021, RxRx1, and JUMP, the method reports competitive or improved endpoint fidelity and mechanism-of-action retention under a unified protocol, plus reduced intermediate support violations on BBBC021.

Significance. If the results hold, the work provides a geometrically grounded formulation for continuous trajectory inference in single-cell perturbation modeling, where endpoint consistency alone is insufficient to ensure meaningful intermediates. The unified evaluation across three datasets and explicit focus on support regularization are strengths that could advance interpretability in this domain.

major comments (2)

[Abstract and §4 (Results)] Abstract and §4 (Results): The central claim that reduced intermediate support violations (via the regularization term) yield biologically interpretable dynamics rests on an untested assumption; the reported metrics are limited to endpoint fidelity, MoA retention, and a geometric proxy (support violations), with no ablation isolating the regularization and no validation of intermediates against independent biological markers or known trajectories.
[§3 (Method)] §3 (Method): The construction of the fixed cellular manifold from instance-segmented representations is presented as sufficient to guarantee constrained, meaningful evolution, but no sensitivity analysis to segmentation quality, alternative representations, or manifold perturbations is provided to support this load-bearing modeling choice.

minor comments (1)

[Abstract] Abstract: The phrase 'competitive or improved' is used without referencing specific baselines, tables, or quantitative deltas; adding these would clarify the strength of the endpoint claims.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments. We address each major point below, providing clarifications and indicating revisions where the manuscript can be strengthened without misrepresenting our current results.

read point-by-point responses

Referee: [Abstract and §4 (Results)] The central claim that reduced intermediate support violations (via the regularization term) yield biologically interpretable dynamics rests on an untested assumption; the reported metrics are limited to endpoint fidelity, MoA retention, and a geometric proxy (support violations), with no ablation isolating the regularization and no validation of intermediates against independent biological markers or known trajectories.

Authors: The support violation metric serves as a direct geometric check that transport remains within the observed single-cell support, which is necessary because multiple stochastic processes can match the same marginals yet produce unsupported intermediates. Our unified protocol shows FreeBridge reduces these violations relative to baselines on BBBC021 while preserving endpoint fidelity and MoA retention. We agree an explicit ablation isolating the regularization term is absent from the current version and will add it. Direct validation against time-resolved biological trajectories is not possible with these endpoint-only assays, which lack paired tracking data; MoA retention provides the available biological corroboration. revision: partial
Referee: [§3 (Method)] The construction of the fixed cellular manifold from instance-segmented representations is presented as sufficient to guarantee constrained, meaningful evolution, but no sensitivity analysis to segmentation quality, alternative representations, or manifold perturbations is provided to support this load-bearing modeling choice.

Authors: Instance segmentation defines the atomic states that constitute the fixed manifold, ensuring transport operates on representations derived from the actual imaged cells rather than an arbitrary latent space. This choice is motivated by the need to respect the empirical support of the data. We acknowledge that the manuscript does not include sensitivity checks on segmentation quality or alternative embeddings. We will add such analyses in the revision, for example by varying segmentation thresholds and comparing feature extractors. revision: yes

standing simulated objections not resolved

Direct validation of inferred intermediate states against independent biological markers or known trajectories, because the datasets provide only separate endpoint marginals without time-series single-cell tracking.

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained via data-driven manifold and regularization.

full rationale

The abstract and description present FreeBridge as introducing a Schrödinger Bridge formulation that defines atomic states from instance-segmented representations to create a fixed manifold, then applies empirical latent support regularization to constrain transport. Reported outcomes (endpoint fidelity, MoA retention, reduced support violations) are empirical results of this construction across external datasets, not reductions of a claimed prediction back to fitted inputs by definition. No equations, self-citations, or uniqueness theorems are shown that would create a load-bearing loop. This is the expected non-circular case for a method paper whose central contribution is the regularization term itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.1-grok · 5732 in / 965 out tokens · 21817 ms · 2026-06-27T14:00:39.354597+00:00 · methodology

0 comments

read the original abstract

High-content imaging assays quantify cellular responses to chemical and genetic perturbations, yet continuous trajectories of individual cells are unobservable because cells are chemically fixed at acquisition. Perturbation modeling therefore reduces to inferring stochastic transport between control and treated populations observed only as separate marginals. While recent generative models achieve strong end-point alignment, boundary consistency does not determine intermediate evolution: multiple stochastic processes may connect identical marginals while traversing regions unsupported by observed single-cell morphologies. We introduce \textbf{FreeBridge}, a Schr\"odinger Bridge formulation for single-cell transition modeling under endpoint-only supervision. FreeBridge defines atomic states as instance-segmented single-cell representations, establishing a fixed cellular manifold, and learns stochastic transport constrained within this geometry via empirical latent support regularization. Across BBBC021, RxRx1, and JUMP, FreeBridge maintains competitive or improved endpoint fidelity and mechanism-of-action retention under a unified evaluation protocol; on BBBC021, it further reduces intermediate support violations. These findings highlight the importance of geometric grounding for biologically interpretable perturbation dynamics. Project page: https://y-research-sbu.github.io/FreeBridge/.

Figures

Figures reproduced from arXiv: 2606.11286 by Chenyu You, Haibin Ling, Jun Ma, Qin Ren, Xurui Wang.

**Figure 1.** Figure 1: Overview of FreeBridge. (a) Multi-cell microscopy images are segmented into atomic single-cell states, forming the State Unit. Each crop is embedded as z = ϕ(x), defining control and perturbed endpoint distributions. On this fixed latent geometry, FreeBridge learns a time-conditioned drift uθ(z, t, c) to model Schrödinger Bridge transport, with an empirical support cost V (z) regularizing intermediate stat… view at source ↗

**Figure 2.** Figure 2: Intermediate trajectories under endpoint-constrained transport. Columns show increasing time along the stochastic evolution. Top (unconstrained): endpoints align, but intermediate states traverse low-support regions. Bottom (FreeBridge): support regularization prevents such excursions and preserves trajectory consistency. With K Euler–Maruyama steps and ∆t = 1/K, the training objective is approximated as … view at source ↗

**Figure 3.** Figure 3: Qualitative comparison of endpoint morphology on BBBC021. Representative single-cell crops for two perturbations are shown. Columns display real targets and samples from IMPA, PhenDiff, CellFlux, and FreeBridge under the unified pipeline. FreeBridge preserves compound-specific structures (e.g., nuclear compaction under Demecolcine and cytoplasmic shrinkage under Mevinolin), whereas baselines show texture … view at source ↗

**Figure 4.** Figure 4: Hyperparameter sensitivity on BBBC021. Performance is evaluated as a function of the support weight λbank, diffusion noise σ, and inference steps (NFE). Results indicate stable endpoint fidelity across a range of settings, with improved semantic retention at moderate regularization strength. 4 Conclusion We introduced FreeBridge, a Schrödinger Bridge formulation for single-cell transition modeling under e… view at source ↗

discussion (0)

Reference graph

Works this paper leans on

25 extracted references · 8 canonical work pages · 5 internal anchors

[1]

Biau, G., Devroye, L.: Lectures on the nearest neighbor method, vol. 246. Springer (2015)

2015
[2]

Demystifying MMD GANs

Bińkowski, M., Sutherland, D.J., Arbel, M., Gretton, A.: Demystifying mmd gans. arXiv preprint arXiv:1801.01401 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018
[3]

In: International Conference on Medical Image Computing and Computer-Assisted Intervention

Bourou, A., Boyer, T., Gheisari, M., Daupin, K., Dubreuil, V., De Thonel, A., Mezger, V., Genovesio, A.: Phendiff: Revealing subtle phenotypes with diffusion models in real images. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 358–367. Springer (2024)

2024
[4]

Cell187(25), 7045–7063 (2024)

Bunne, C., Roohani, Y., Rosen, Y., Gupta, A., Zhang, X., Roed, M., Alexandrov, T., AlQuraishi, M., Brennan, P., Burkhardt, D.B., et al.: How to build the virtual cell with artificial intelligence: Priorities and opportunities. Cell187(25), 7045–7063 (2024)

2024
[5]

Nature Chemical Biology3(8), 461–465 (2007)

Carpenter, A.E.: Image-based chemical screening. Nature Chemical Biology3(8), 461–465 (2007)

2007
[6]

BioRxiv pp

Chandrasekaran, S.N., Ackerman, J., Alix, E., Ando, D.M., Arevalo, J., Bennion, M., Boisseau, N., Borowa, A., Boyd, J.D., Brino, L., et al.: Jump cell painting dataset: morphological impact of 136,000 chemical and genetic perturbations. BioRxiv pp. 2023–03 (2023)

2023
[7]

Journal of Optimization Theory and Applications169(2), 671–691 (2016)

Chen, Y., Georgiou, T.T., Pavon, M.: On the relation between optimal transport and schrödinger bridges: A stochastic control viewpoint. Journal of Optimization Theory and Applications169(2), 671–691 (2016)

2016
[8]

Advances in neural information processing systems30(2017)

Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems30(2017)

2017
[9]

Advances in neural information processing systems33, 6840–6851 (2020)

Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

2020
[10]

Adam: A Method for Stochastic Optimization

Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[11]

arXiv preprint arXiv:1308.0215 , title =

Léonard, C.: A survey of the schrödinger problem and some of its connections with optimal transport. arXiv preprint arXiv:1308.0215 (2013)

work page arXiv 2013
[12]

Flow Matching for Generative Modeling

Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022
[13]

In: International Conference on Learning Representations

Liu, G.H., Lipman, Y., Nickel, M., Karrer, B., Theodorou, E., Chen, R.T.: Gen- eralized schrödinger bridge matching. In: International Conference on Learning Representations. vol. 2024, pp. 14527–14552 (2024)

2024
[14]

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

Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022
[15]

Nature methods9(7), 637 (2012)

Ljosa, V., Sokolnicki, K.L., Carpenter, A.E.: Annotated high-throughput microscopy image sets for validation. Nature methods9(7), 637 (2012)

2012
[16]

Nature Communications16(1), 505 (2025)

Palma, A., Theis, F.J., Lotfollahi, M.: Predicting cell morphological responses to perturbations using generative modeling. Nature Communications16(1), 505 (2025)

2025
[17]

arXiv preprint arXiv:2511.19917 (2025)

Ren, Q., Wang, Y., Guo, L., Zhang, W., Fan, Z., You, C.: Scale where it matters: Training-free localized scaling for diffusion models. arXiv preprint arXiv:2511.19917 (2025) FreeBridge: Variational Schrödinger Bridges 11

work page arXiv 2025
[18]

Routledge (2018)

Silverman, B.W.: Density estimation for statistics and data analysis. Routledge (2018)

2018
[19]

Trends in cell biology13(11), 570–576 (2003)

Slepchenko, B.M., Schaff, J.C., Macara, I., Loew, L.M.: Quantitative cell biology with the virtual cell. Trends in cell biology13(11), 570–576 (2003)

2003
[20]

Score-Based Generative Modeling through Stochastic Differential Equations

Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score- based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 (2020)

work page internal anchor Pith review Pith/arXiv arXiv 2011
[21]

Nature methods18(1), 100–106 (2021)

Stringer, C., Wang, T., Michaelos, M., Pachitariu, M.: Cellpose: a generalist algo- rithm for cellular segmentation. Nature methods18(1), 100–106 (2021)

2021
[22]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

Sun, S., Wang, Y., Zhang, H., Xiong, Y., Ren, Q., Fang, R., Xie, X., You, C.: Ouroboros: Single-step diffusion models for cycle-consistent forward and inverse rendering. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

2025
[23]

In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (2023)

Sypetkowski, M., Rezanejad, M., Saberian, S., Kraus, O., Urbanik, J., Taylor, J., Mabey, B., Victors, M., Yosinski, J., Sereshkeh, A.R., et al.: Rxrx1: A dataset for evaluating experimental batch correction methods. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (2023)

2023
[24]

In: International conference on machine learning (2026)

Wang, Y., Ma, Y., Li, W., You, C.: Let eeg models learn eeg. In: International conference on machine learning (2026)

2026
[25]

arXiv preprint arXiv:2502.09775 (2025)

Zhang, Y., Su, Y., Wang, C., Li, T., Wefers, Z., Nirschl, J., Burgess, J., Ding, D., Lozano, A., Lundberg, E., et al.: Cellflux: Simulating cellular morphology changes via flow matching. arXiv preprint arXiv:2502.09775 (2025)

work page arXiv 2025

[1] [1]

Biau, G., Devroye, L.: Lectures on the nearest neighbor method, vol. 246. Springer (2015)

2015

[2] [2]

Demystifying MMD GANs

Bińkowski, M., Sutherland, D.J., Arbel, M., Gretton, A.: Demystifying mmd gans. arXiv preprint arXiv:1801.01401 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018

[3] [3]

In: International Conference on Medical Image Computing and Computer-Assisted Intervention

Bourou, A., Boyer, T., Gheisari, M., Daupin, K., Dubreuil, V., De Thonel, A., Mezger, V., Genovesio, A.: Phendiff: Revealing subtle phenotypes with diffusion models in real images. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 358–367. Springer (2024)

2024

[4] [4]

Cell187(25), 7045–7063 (2024)

Bunne, C., Roohani, Y., Rosen, Y., Gupta, A., Zhang, X., Roed, M., Alexandrov, T., AlQuraishi, M., Brennan, P., Burkhardt, D.B., et al.: How to build the virtual cell with artificial intelligence: Priorities and opportunities. Cell187(25), 7045–7063 (2024)

2024

[5] [5]

Nature Chemical Biology3(8), 461–465 (2007)

Carpenter, A.E.: Image-based chemical screening. Nature Chemical Biology3(8), 461–465 (2007)

2007

[6] [6]

BioRxiv pp

Chandrasekaran, S.N., Ackerman, J., Alix, E., Ando, D.M., Arevalo, J., Bennion, M., Boisseau, N., Borowa, A., Boyd, J.D., Brino, L., et al.: Jump cell painting dataset: morphological impact of 136,000 chemical and genetic perturbations. BioRxiv pp. 2023–03 (2023)

2023

[7] [7]

Journal of Optimization Theory and Applications169(2), 671–691 (2016)

Chen, Y., Georgiou, T.T., Pavon, M.: On the relation between optimal transport and schrödinger bridges: A stochastic control viewpoint. Journal of Optimization Theory and Applications169(2), 671–691 (2016)

2016

[8] [8]

Advances in neural information processing systems30(2017)

Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems30(2017)

2017

[9] [9]

Advances in neural information processing systems33, 6840–6851 (2020)

Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

2020

[10] [10]

Adam: A Method for Stochastic Optimization

Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014

[11] [11]

arXiv preprint arXiv:1308.0215 , title =

Léonard, C.: A survey of the schrödinger problem and some of its connections with optimal transport. arXiv preprint arXiv:1308.0215 (2013)

work page arXiv 2013

[12] [12]

Flow Matching for Generative Modeling

Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022

[13] [13]

In: International Conference on Learning Representations

Liu, G.H., Lipman, Y., Nickel, M., Karrer, B., Theodorou, E., Chen, R.T.: Gen- eralized schrödinger bridge matching. In: International Conference on Learning Representations. vol. 2024, pp. 14527–14552 (2024)

2024

[14] [14]

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

Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022

[15] [15]

Nature methods9(7), 637 (2012)

Ljosa, V., Sokolnicki, K.L., Carpenter, A.E.: Annotated high-throughput microscopy image sets for validation. Nature methods9(7), 637 (2012)

2012

[16] [16]

Nature Communications16(1), 505 (2025)

Palma, A., Theis, F.J., Lotfollahi, M.: Predicting cell morphological responses to perturbations using generative modeling. Nature Communications16(1), 505 (2025)

2025

[17] [17]

arXiv preprint arXiv:2511.19917 (2025)

Ren, Q., Wang, Y., Guo, L., Zhang, W., Fan, Z., You, C.: Scale where it matters: Training-free localized scaling for diffusion models. arXiv preprint arXiv:2511.19917 (2025) FreeBridge: Variational Schrödinger Bridges 11

work page arXiv 2025

[18] [18]

Routledge (2018)

Silverman, B.W.: Density estimation for statistics and data analysis. Routledge (2018)

2018

[19] [19]

Trends in cell biology13(11), 570–576 (2003)

Slepchenko, B.M., Schaff, J.C., Macara, I., Loew, L.M.: Quantitative cell biology with the virtual cell. Trends in cell biology13(11), 570–576 (2003)

2003

[20] [20]

Score-Based Generative Modeling through Stochastic Differential Equations

Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score- based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 (2020)

work page internal anchor Pith review Pith/arXiv arXiv 2011

[21] [21]

Nature methods18(1), 100–106 (2021)

Stringer, C., Wang, T., Michaelos, M., Pachitariu, M.: Cellpose: a generalist algo- rithm for cellular segmentation. Nature methods18(1), 100–106 (2021)

2021

[22] [22]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

Sun, S., Wang, Y., Zhang, H., Xiong, Y., Ren, Q., Fang, R., Xie, X., You, C.: Ouroboros: Single-step diffusion models for cycle-consistent forward and inverse rendering. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

2025

[23] [23]

In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (2023)

Sypetkowski, M., Rezanejad, M., Saberian, S., Kraus, O., Urbanik, J., Taylor, J., Mabey, B., Victors, M., Yosinski, J., Sereshkeh, A.R., et al.: Rxrx1: A dataset for evaluating experimental batch correction methods. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (2023)

2023

[24] [24]

In: International conference on machine learning (2026)

Wang, Y., Ma, Y., Li, W., You, C.: Let eeg models learn eeg. In: International conference on machine learning (2026)

2026

[25] [25]

arXiv preprint arXiv:2502.09775 (2025)

Zhang, Y., Su, Y., Wang, C., Li, T., Wefers, Z., Nirschl, J., Burgess, J., Ding, D., Lozano, A., Lundberg, E., et al.: Cellflux: Simulating cellular morphology changes via flow matching. arXiv preprint arXiv:2502.09775 (2025)

work page arXiv 2025