arxiv: 2605.03573 · v2 · submitted 2026-05-05 · 📊 stat.ML · cs.LG

Recognition: 1 theorem link

· Lean Theorem

Stochastic Schr\"odinger Diffusion Models for Pure-State Ensemble Generation

Jian Xu , Wei Chen. Chao Li , Jingyuan Zheng , Delu Zeng , John Paisley , Qibin Zhao

Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:08 UTC · model grok-4.3

classification 📊 stat.ML cs.LG

keywords diffusion modelsquantum machine learningpure statesscore-based generationFubini-Study metricstochastic Schrödinger equationRiemannian diffusiondata augmentation

0 comments

The pith

Stochastic Schrödinger diffusion models generate new quantum pure states from target ensembles on the complex projective space.

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

The paper introduces SSDMs as a score-based generative method that operates directly on quantum pure-state ensembles encoded in the complex projective space equipped with the Fubini-Study metric. It formulates a forward diffusion process via a stochastic Schrödinger equation and derives the corresponding reverse-time dynamics using the Riemannian score, bypassing the need for analytic transition densities through a local Euclidean approximation. A sympathetic reader would care because this enables sampling new quantum representations at the state level rather than perturbing classical inputs, which in turn supports data augmentation for quantum machine learning models. The central mechanism rests on mapping an Ornstein-Uhlenbeck teacher score back onto the manifold while preserving the intrinsic geometry.

Core claim

SSDMs realize Riemannian diffusion on CP^{d-1} through a stochastic Schrödinger equation whose reverse process is driven by the Fubini-Study score; training proceeds via a local-time objective that employs an Ornstein-Uhlenbeck approximation in Fubini-Study normal coordinates to supply an analytic teacher score that is then lifted back to the manifold.

What carries the argument

The Riemannian score on the Fubini-Study manifold, approximated locally by an Ornstein-Uhlenbeck process in normal coordinates and mapped back to drive the reverse stochastic Schrödinger dynamics.

If this is right

SSDM samples reproduce observable moments, overlap-kernel MMD distances, and entanglement measures of the target pure-state ensemble.
Representation-level augmentation with SSDM-generated states improves generalization performance on downstream quantum machine learning tasks.
The method permits direct sampling from pure-state distributions without repeated preparation from perturbed classical data.

Where Pith is reading between the lines

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

If the local approximation remains stable at higher dimensions, SSDMs could serve as a general tool for manifold-valued score matching beyond quantum states.
The same local-time objective might be adapted to other Riemannian manifolds where exact transition densities are unavailable.
Combining SSDM augmentation with existing variational quantum algorithms could test whether representation-level diversity reduces barren-plateau effects.

Load-bearing premise

The local Euclidean Ornstein-Uhlenbeck approximation in Fubini-Study normal coordinates supplies a teacher score accurate enough that mapping it back to the manifold introduces no significant bias in the learned reverse dynamics.

What would settle it

Compare the overlap-kernel MMD or entanglement entropy of states generated by the trained SSDM against the same quantities obtained from exact manifold sampling; a statistically significant mismatch that grows with dimension would indicate the approximation introduces unacceptable bias.

Figures

Figures reproduced from arXiv: 2605.03573 by Delu Zeng, Jian Xu, Jingyuan Zheng, John Paisley, Qibin Zhao, Wei Chen. Chao Li.

**Figure 1.** Figure 1: Illustration of the forward diffusion on the quantum pure-state manifold and its reverse-time generative process driven by a learned Riemannian score. coordinates or converted to Ito form, additional geometry- ˆ dependent correction terms appear (e.g., Levi-Civita connection and divergence terms associated with the Riemannian volume measure). In our implementation, we perform updates in local orthonormal… view at source ↗

**Figure 2.** Figure 2: Training stability across system sizes n ∈ {2, 4, 6} (rows). Columns report training loss, F0, MMD, ∆obs, and entropic Wasserstein distance (Ent. W1). Schrodinger (SSE) realization, and (ii) learning a ¨ Riemannian score field under the Fubini–Study (FS) geometry. A key practical challenge—the intractability of transition densities on curved manifolds—is addressed via a local-time learning signal derived f… view at source ↗

read the original abstract

In quantum machine learning (QML), classical data are often encoded as quantum pure states and processed directly as quantum representations, motivating representation-level generative modeling that samples new quantum states from an underlying pure-state ensemble rather than re-preparing them from perturbed classical inputs. However, extending \emph{score-based} diffusion models with well-defined reverse-time samplers to quantum pure-state ensembles remains challenging, due to the non-Euclidean geometry of the complex projective space $\mathbb{CP}^{d-1}$ and the intractability of transition densities. We propose \emph{Stochastic Schr\"odinger Diffusion Models} (SSDMs), an intrinsic score-based generative framework on $\mathbb{CP}^{d-1}$ endowed with the Fubini--Study (FS) metric. SSDMs formulate a forward Riemannian diffusion with a stochastic Schr\"odinger equation (SSE) realization, and derive reverse-time dynamics driven by the Riemannian score $\nabla_{\mathrm{FS}} \log p_t$. To enable training without analytic transition densities, we introduce a local-time objective based on a local Euclidean Ornstein--Uhlenbeck approximation in FS normal coordinates, yielding an analytic teacher score mapped back to the manifold. Experiments show that SSDMs faithfully capture target pure-state ensemble statistics, including observable moments, overlap-kernel MMD, and entanglement measures, and that SSDM-generated quantum representations improve downstream QML generalization via representation-level data augmentation.

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.

Desk Editor's Note private letter to a colleague

SSDMs give an intrinsic diffusion process on the Fubini-Study manifold for pure quantum states, but the local OU approximation used for training is the part that needs checking.

read the letter

The paper introduces Stochastic Schrödinger Diffusion Models that run a forward diffusion directly on CP^{d-1} using a stochastic Schrödinger equation and then learn a reverse process driven by the Riemannian score under the Fubini-Study metric. That combination is new relative to prior Euclidean diffusion work and to existing quantum generative models that usually embed or flatten the states first. The setup respects the geometry from the start, which is the main technical step forward. Experiments report that samples match the target ensemble on observable moments, overlap-kernel MMD, and entanglement measures, and that the generated states give a modest lift in downstream QML tasks when used for representation-level augmentation. Those results are presented clearly enough to show the framework is at least workable in practice. The soft spot is the training objective. They replace the intractable transition density with a local Euclidean Ornstein-Uhlenbeck approximation inside Fubini-Study normal coordinates and then map the resulting score back to the tangent space. If the chart radius is not small compared with the curvature scale or if the data ensemble spreads away from the origin, that first-order approximation can distort the score and push the learned reverse dynamics toward a different measure. The abstract and reported numbers do not include error bounds or sensitivity checks on this choice, so it is hard to tell how much the good empirical numbers depend on the approximation holding. The stress-test concern therefore lands on the central claim. This is aimed at researchers who already work on manifold-valued diffusion or on quantum representations that need to be sampled rather than re-encoded from classical data. A reader who wants to try geometry-aware generative modeling in QML would get a usable starting point and could test the approximation themselves. The paper shows clear thinking about the geometry and the training workaround, so it deserves a serious referee even though the approximation step will probably require more justification or tighter analysis in revision.

Referee Report

3 major / 2 minor

Summary. The paper introduces Stochastic Schrödinger Diffusion Models (SSDMs) as an intrinsic score-based generative framework for sampling from pure-state ensembles on the complex projective space CP^{d-1} equipped with the Fubini-Study metric. It defines a forward Riemannian diffusion via a stochastic Schrödinger equation realization, derives the corresponding reverse-time SDE driven by the Riemannian score ∇_{FS} log p_t, and enables training via a local-time objective that employs a local Euclidean Ornstein-Uhlenbeck approximation in Fubini-Study normal coordinates to produce an analytic teacher score subsequently mapped back to the manifold. Experiments are reported to show that SSDM samples reproduce target ensemble statistics (observable moments, overlap-kernel MMD, entanglement measures) and that the generated states improve downstream quantum machine learning generalization when used for representation-level data augmentation.

Significance. If the local approximation is shown to be sufficiently accurate, the work would provide a principled way to perform score-based diffusion directly on quantum state manifolds, addressing a gap in representation-level generative modeling for QML. The coherent formulation of the forward SSE and reverse dynamics, together with the reported fidelity on multiple quantum statistics, would constitute a meaningful technical contribution to manifold-valued generative models.

major comments (3)

[Training objective / local approximation] The section deriving the training objective (local Euclidean Ornstein-Uhlenbeck approximation in FS normal coordinates): the central claim that the learned reverse dynamics faithfully reproduce the target measure rests on this approximation introducing negligible bias when the score is mapped back to the tangent space. No error bounds, curvature-scale analysis, or sensitivity study with respect to chart radius is provided, leaving open the possibility that the first-order local approximation systematically distorts the score for ensembles with non-negligible support away from the chart origin.
[Experiments] Experimental results on observable moments, overlap-kernel MMD, and entanglement measures: these quantities are reported as faithfully captured, yet without quantitative comparison to a baseline that uses the exact Riemannian score (or a higher-order approximation) it is impossible to isolate whether residual discrepancies arise from the local approximation itself.
[Downstream QML experiments] Downstream QML augmentation experiments: the claimed generalization improvement is attributed to SSDM-generated states, but the manuscript does not report controls that hold the representation distribution fixed while varying only the generative mechanism, making it difficult to attribute gains specifically to the manifold-aware reverse dynamics.

minor comments (2)

[Notation / derivation of reverse dynamics] Notation for the mapping of the Euclidean score back to the FS tangent space should be made fully explicit, including the precise identification of the normal coordinates and the projection operator used.
[Implementation details] The diffusion schedule and noise parameters are listed as free; a brief discussion of their selection procedure and sensitivity would improve reproducibility.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the thoughtful comments and suggestions. We believe the manuscript can be strengthened by addressing the concerns regarding the local approximation and experimental validation. We respond to each major comment below.

read point-by-point responses

Referee: The section deriving the training objective (local Euclidean Ornstein-Uhlenbeck approximation in FS normal coordinates): the central claim that the learned reverse dynamics faithfully reproduce the target measure rests on this approximation introducing negligible bias when the score is mapped back to the tangent space. No error bounds, curvature-scale analysis, or sensitivity study with respect to chart radius is provided, leaving open the possibility that the first-order local approximation systematically distorts the score for ensembles with non-negligible support away from the chart origin.

Authors: We appreciate this observation. The local approximation is chosen for tractability, but we agree that its accuracy should be better characterized. In the revised manuscript, we will incorporate a curvature-based error analysis for the Fubini-Study manifold and conduct sensitivity experiments by varying the normal coordinate chart radius to assess the impact on the learned score. revision: yes
Referee: Experimental results on observable moments, overlap-kernel MMD, and entanglement measures: these quantities are reported as faithfully captured, yet without quantitative comparison to a baseline that uses the exact Riemannian score (or a higher-order approximation) it is impossible to isolate whether residual discrepancies arise from the local approximation itself.

Authors: We agree that such a comparison would be valuable for isolating the approximation's effect. However, the exact Riemannian score is intractable, which is precisely why the local-time objective with the Ornstein-Uhlenbeck approximation was introduced. We cannot provide this baseline. We will instead add comparisons to Euclidean diffusion models and other manifold generative approaches, along with a discussion of the approximation's limitations. revision: no
Referee: Downstream QML augmentation experiments: the claimed generalization improvement is attributed to SSDM-generated states, but the manuscript does not report controls that hold the representation distribution fixed while varying only the generative mechanism, making it difficult to attribute gains specifically to the manifold-aware reverse dynamics.

Authors: We thank the referee for this suggestion. To better isolate the contribution of the manifold-aware dynamics, we will add control experiments in the revised manuscript. These will include using the same set of generated states but trained with different objectives, and comparisons where the generative mechanism is varied while keeping the target distribution fixed. revision: yes

standing simulated objections not resolved

Quantitative comparison to a baseline using the exact Riemannian score, which is intractable.

Circularity Check

0 steps flagged

Derivation chain self-contained; no circular reductions identified

full rationale

The forward Riemannian diffusion is realized via stochastic Schrödinger equation and the reverse dynamics are derived directly from the Riemannian score ∇_FS log p_t on CP^{d-1}. The training objective employs an external local Euclidean Ornstein-Uhlenbeck approximation in Fubini-Study normal coordinates as a modeling device to obtain an analytic teacher score; this approximation does not reduce the claimed fidelity on moments, MMD or entanglement to any fitted input by construction. No self-citations are load-bearing for the central claims, no ansatz is smuggled, and no uniqueness theorem is invoked from prior author work. The derivation remains independent of the target statistics.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on standard Riemannian geometry and stochastic calculus; the local-time objective introduces an approximation whose validity is not independently verified in the provided text.

free parameters (1)

diffusion schedule and noise parameters
Typical hyperparameters in score-based diffusion models that must be chosen or tuned for the Riemannian setting.

axioms (2)

standard math The Fubini-Study metric is the appropriate Riemannian structure on CP^{d-1} for quantum pure states.
Invoked throughout the geometric formulation of the diffusion process.
ad hoc to paper The local Euclidean Ornstein-Uhlenbeck process in normal coordinates yields a usable approximation to the true Riemannian score.
Central to enabling training without analytic transition densities.

pith-pipeline@v0.9.0 · 5570 in / 1323 out tokens · 37439 ms · 2026-05-12T04:08:00.635346+00:00 · methodology

Review history (4 revisions) →

discussion (0)

Reference graph

Works this paper leans on

70 extracted references · 70 canonical work pages

[1]

2022 , eprint=

Denoising Diffusion Implicit Models , author=. 2022 , eprint=

work page 2022
[2]

Physical Review Letters , volume=

Generative quantum machine learning via denoising diffusion probabilistic models , author=. Physical Review Letters , volume=. 2024 , publisher=

work page 2024
[3]

Physical review letters , volume=

Quantum generative adversarial learning , author=. Physical review letters , volume=. 2018 , publisher=

work page 2018
[4]

International Conference on Learning Representations , year=

Score-Based Generative Modeling through Stochastic Differential Equations , author=. International Conference on Learning Representations , year=

work page
[5]

Advances in neural information processing systems , volume=

Riemannian score-based generative modelling , author=. Advances in neural information processing systems , volume=

work page
[6]

International Conference on Artificial Intelligence and Statistics , pages=

Adaptivity of diffusion models to manifold structures , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2024 , organization=

work page 2024
[7]

Aip advances , volume=

A short introduction to the Lindblad master equation , author=. Aip advances , volume=. 2020 , publisher=

work page 2020
[8]

Proceedings of the AAAI Symposium Series , volume=

Quantum Diffusion Model for Quark and Gluon Jet Generation , author=. Proceedings of the AAAI Symposium Series , volume=

work page
[9]

Journal of Physics A: Mathematical and Theoretical , volume=

The generalized Gell--Mann representation and violation of the CHSH inequality by a general two-qudit state , author=. Journal of Physics A: Mathematical and Theoretical , volume=. 2020 , publisher=

work page 2020
[10]

Annales de l'IHP Probabilit

Stratonovich stochastic differential equations driven by general semimartingales , author=. Annales de l'IHP Probabilit

work page
[11]

Advances in Neural Information Processing Systems , volume=

Scaling Riemannian diffusion models , author=. Advances in Neural Information Processing Systems , volume=

work page
[12]

Stochastic schr

Bouten, Luc and Guta, Madalin and Maassen, Hans , journal=. Stochastic schr. 2004 , publisher=

work page 2004
[13]

Advances in Neural Information Processing Systems , editor=

Riemannian Diffusion Models , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

work page 2022
[14]

Advances in Neural Information Processing Systems , editor=

Riemannian Score-Based Generative Modelling , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

work page 2022
[15]

Physical Review A , volume=

Differentiable learning of quantum circuit born machines , author=. Physical Review A , volume=. 2018 , publisher=

work page 2018
[16]

npj Quantum Information , volume=

The Born supremacy: quantum advantage and training of an Ising Born machine , author=. npj Quantum Information , volume=. 2020 , publisher=

work page 2020
[17]

npj Quantum information , volume=

A generative modeling approach for benchmarking and training shallow quantum circuits , author=. npj Quantum information , volume=. 2019 , publisher=

work page 2019
[18]

Physical Review X , volume=

Quantum boltzmann machine , author=. Physical Review X , volume=. 2018 , publisher=

work page 2018
[19]

Physical Review A , year=

Tomography and generative training with quantum Boltzmann machines , author=. Physical Review A , year=

work page
[20]

Quantum Machine Intelligence , year=

Variational quantum Boltzmann machines , author=. Quantum Machine Intelligence , year=

work page
[21]

Uncertainty in Artificial Intelligence Conference , pages =

Sliced Score Matching: A Scalable Approach to Density and Score Estimation , author =. Uncertainty in Artificial Intelligence Conference , pages =. 2020 , editor =

work page 2020
[22]

The Annals of Probability , pages=

Time reversal of diffusions , author=. The Annals of Probability , pages=. 1986 , publisher=

work page 1986
[23]

Stochastic Processes and their Applications , volume=

Reverse-time diffusion equation models , author=. Stochastic Processes and their Applications , volume=. 1982 , publisher=

work page 1982
[24]

Advances in Neural Information Processing Systems , volume=

Denoising diffusion probabilistic models , author=. Advances in Neural Information Processing Systems , volume=

work page
[25]

2024 , eprint=

Optical Diffusion Models for Image Generation , author=. 2024 , eprint=

work page 2024
[26]

2021 , eprint=

Diffusion Models Beat GANs on Image Synthesis , author=. 2021 , eprint=

work page 2021
[27]

G., Onodera, T., Stein, M

Wright, Logan G. and Onodera, Tatsuhiro and Stein, Martin M. and Wang, Tianyu and Schachter, Darren T. and Hu, Zoey and McMahon, Peter L. , year=. Deep physical neural networks trained with backpropagation , volume=. Nature , publisher=. doi:10.1038/s41586-021-04223-6 , number=

work page doi:10.1038/s41586-021-04223-6
[28]

2024 , eprint=

Experimental quantum-enhanced kernels on a photonic processor , author=. 2024 , eprint=

work page 2024
[29]

2021 , eprint=

Score-Based Generative Modeling through Stochastic Differential Equations , author=. 2021 , eprint=

work page 2021
[30]

2023 , eprint=

Denoising Diffusion Bridge Models , author=. 2023 , eprint=

work page 2023
[31]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

work page 2022
[32]

2023 , eprint=

Flow Matching for Generative Modeling , author=. 2023 , eprint=

work page 2023
[33]

, author=

Quantum Generative Adversarial Learning. , author=. Physical review letters , year=

work page
[34]

arXiv preprint arXiv:1804.08641 , year=

Quantum generative adversarial networks , author=. arXiv preprint arXiv:1804.08641 , year=

work page arXiv
[35]

npj Quantum Information , year=

Quantum Generative Adversarial Networks for learning and loading random distributions , author=. npj Quantum Information , year=

work page
[36]

arXiv preprint arXiv:2401.07039 , year=

Quantum generative diffusion model: a fully quantum-mechanical model for generating quantum state ensemble , author=. arXiv preprint arXiv:2401.07039 , year=

work page arXiv
[37]

arXiv preprint arXiv:2511.12221 , year=

Channel-Constrained Markovian Quantum Diffusion Model from Open System Perspective , author=. arXiv preprint arXiv:2511.12221 , year=

work page arXiv
[38]

Advanced Quantum Technologies , volume=

Quantum-noise-driven generative diffusion models , author=. Advanced Quantum Technologies , volume=. 2025 , publisher=

work page 2025
[39]

, author=

Wave-function approach to dissipative processes in quantum optics. , author=. Physical review letters , year=

work page
[40]

Journal of Physics A , year=

The quantum-state diffusion model applied to open systems , author=. Journal of Physics A , year=

work page
[41]

Classical and Quantum Gravity , year=

Quantum Measurement and Control , author=. Classical and Quantum Gravity , year=

work page
[42]

Quantum Machine Intelligence , year=

Quantum latent diffusion models , author=. Quantum Machine Intelligence , year=

work page
[43]

IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium , year=

Leveraging Quantum Latent Diffusion Models for Data Augmentation on The Eurosat Dataset , author=. IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium , year=

work page 2025
[44]

2024 IEEE International Conference on Quantum Software (QSW) , pages=

Quantum denoising diffusion models , author=. 2024 IEEE International Conference on Quantum Software (QSW) , pages=. 2024 , organization=

work page 2024
[45]

Physical review letters , year=

Generative quantum machine learning via denoising diffusion probabilistic models , author=. Physical review letters , year=

work page
[46]

Physical Review A , volume=

Mixed-state quantum denoising diffusion probabilistic model , author=. Physical Review A , volume=. 2025 , publisher=

work page 2025
[47]

Physical Review A , volume=

Stochastic unraveling of positive quantum dynamics , author=. Physical Review A , volume=. 2017 , publisher=

work page 2017
[48]

Physical Review E , volume=

Stochastic unraveling of time-local quantum master equations beyond the Lindblad class , author=. Physical Review E , volume=. 2002 , publisher=

work page 2002
[49]

Physical Review A , volume=

Unraveling quantum environments: Transformer-assisted learning in Lindblad dynamics , author=. Physical Review A , volume=. 2025 , publisher=

work page 2025
[50]

2025 , url=

Shigui Li and Wei Chen and Delu Zeng , booktitle=. 2025 , url=

work page 2025
[51]

ACM Computing Surveys , year=

Diffusion Models: A Comprehensive Survey of Methods and Applications , author=. ACM Computing Surveys , year=

work page
[52]

International Conference on Machine Learning , pages=

Equivariant diffusion for molecule generation in 3d , author=. International Conference on Machine Learning , pages=. 2022 , organization=

work page 2022
[53]

International Conference on Learning Representations , year=

GeoDiff: A Geometric Diffusion Model for Molecular Conformation Generation , author=. International Conference on Learning Representations , year=

work page
[54]

International Conference on Learning Representations , year=

WaveGrad: Estimating Gradients for Waveform Generation , author=. International Conference on Learning Representations , year=

work page
[55]

International Conference on Learning Representations , year=

DiffWave: A Versatile Diffusion Model for Audio Synthesis , author=. International Conference on Learning Representations , year=

work page
[56]

Nature , volume=

De novo design of protein structure and function with RFdiffusion , author=. Nature , volume=. 2023 , publisher=

work page 2023
[57]

The Eleventh International Conference on Learning Representations , year=

Diffusion Probabilistic Modeling of Protein Backbones in 3D for the motif-scaffolding problem , author=. The Eleventh International Conference on Learning Representations , year=

work page
[58]

Nature communications , volume=

Protein structure generation via folding diffusion , author=. Nature communications , volume=. 2024 , publisher=

work page 2024
[59]

Nature , volume=

Quantum machine learning , author=. Nature , volume=. 2017 , publisher=

work page 2017
[60]

Quantum , volume=

Quantum computing in the NISQ era and beyond , author=. Quantum , volume=. 2018 , publisher=

work page 2018
[61]

Physical review letters , volume=

Quantum machine learning in feature Hilbert spaces , author=. Physical review letters , volume=. 2019 , publisher=

work page 2019
[62]

Nature , volume=

Supervised learning with quantum-enhanced feature spaces , author=. Nature , volume=. 2019 , publisher=

work page 2019
[63]

Nature communications , volume=

Power of data in quantum machine learning , author=. Nature communications , volume=. 2021 , publisher=

work page 2021
[64]

Communications in Mathematical Physics , volume=

Geometry of quantum states , author=. Communications in Mathematical Physics , volume=. 1968 , publisher=

work page 1968
[65]

Thirty-seventh Conference on Neural Information Processing Systems , year=

Scaling Riemannian Diffusion Models , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

work page
[66]

2002 , publisher=

Stochastic analysis on manifolds , author=. 2002 , publisher=

work page 2002
[67]

Physical Review A—Atomic, Molecular, and Optical Physics , volume=

Exact and approximate unitary 2-designs and their application to fidelity estimation , author=. Physical Review A—Atomic, Molecular, and Optical Physics , volume=. 2009 , publisher=

work page 2009
[68]

Physical Review A , volume=

Multiqubit Clifford groups are unitary 3-designs , author=. Physical Review A , volume=. 2017 , publisher=

work page 2017
[69]

The journal of machine learning research , volume=

A kernel two-sample test , author=. The journal of machine learning research , volume=. 2012 , publisher=

work page 2012
[70]

International Conference on Machine Learning , year=

Dequantified Diffusion-Schr\"odinger Bridge for Density Ratio Estimation , author=. International Conference on Machine Learning , year=

work page