Pith. sign in

REVIEW 3 major objections 6 minor 87 references

Stop the diffusion ODE early and decode the clean sample; affine paths already contain the endpoint.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 16:06 UTC pith:U2AOOUPI

load-bearing objection Clean algebraic formalization of early-exit x0 decoding that actually delivers 20-70% NFE cuts on off-the-shelf checkpoints, with theory that holds up. the 3 major comments →

arxiv 2607.06114 v2 pith:U2AOOUPI submitted 2026-07-07 cs.LG cs.AI

x-Prediction Is All You Need:Training-Free Accelerated Generation via Endpoint Decodability

Xin Peng , Ang Gao This is my paper
classification cs.LG cs.AI
keywords endpoint decodabilitytruncated jump samplingdiffusion modelsflow matchingtraining-free accelerationx-predictionaffine probability pathsODE sampling
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.

Diffusion and flow-matching models generate high-quality images, but their ordinary differential equation samplers still burn tens to hundreds of neural function evaluations. Most accelerators retrain, distill, or straighten trajectories. This paper shows that every standard affine probability path already exposes the clean endpoint: from any intermediate state and its path velocity one can recover a closed-form estimate of the clean sample. The authors call the property endpoint decodability and prove that the algebraic decoder is exactly the minimum-mean-square-error estimator of the clean sample under ordinary training. Truncated Jump Sampling simply integrates a fraction of the usual steps, applies the decoder once, and returns the result. Across large text-to-image checkpoints and class-conditional benchmarks the method cuts function evaluations by 20–70 percent while preserving essentially full quality, with no retraining or architecture change. The same analysis shows that straight trajectories are sufficient but unnecessary for this acceleration.

Core claim

For any non-degenerate affine probability path the intermediate state and path velocity uniquely determine the clean sample through a closed-form decoder, and under the standard ℓ2 training objective that decoder is precisely the minimum-mean-square-error estimator of the clean sample. Consequently one may safely stop the ODE early and jump to the decoded endpoint without redesigning the trajectory.

What carries the argument

Endpoint decodability: the path determinant Δt = α̇t σt − αt σ̇t is nonzero, so the linear system relating (xt, ut) to (x0, ε) is invertible and the decoder x0 = (σt ut − σ̇t xt)/Δt recovers the clean sample; under optimal velocity prediction this equals E[x0|xt]. Truncated Jump Sampling is the resulting early-exit algorithm.

Load-bearing premise

The pretrained model must already be accurate enough that its estimation error is negligible once the intermediate state carries most of the endpoint information; if that residual model error stays large, early exit fails even though the algebra is exact.

What would settle it

On a well-trained checkpoint, measure whether quality metrics remain monotonic and near full-ODE levels when the ODE is stopped at the paper’s recommended early-exit fractions; a sharp quality collapse at those fractions, or a large residual error between the decoded endpoint and the true clean sample once the irreducible uncertainty has decayed, would falsify the claim.

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

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper formalizes endpoint decodability for affine probability paths: whenever the path determinant Δt ≠ 0, the pair (xt, ut) recovers x0 by the closed-form decoder of Theorem 5, and under the standard ℓ2 training objective this decoder is exactly the MMSE estimator E[x0|xt] (Theorem 6). From this they propose Truncated Jump Sampling (TJS): integrate the pretrained ODE only to an early-exit time t* and return the decoded x0. Error analysis (Theorem 9) decomposes TJS MSE into model error plus irreducible endpoint uncertainty U(t*), with no curvature term; Proposition 12 shows straight trajectories are sufficient but not necessary. Experiments on SDXL, SD3.5M, Z-Image-Turbo, ImageNet-256, CIFAR-10, and MNIST report 20–70% NFE reductions at near-matched quality, with sampler/schedule/CFG/step-count ablations and multi-metric T2I sweeps.

Significance. If the claims hold, the work supplies a genuinely training-free acceleration route that applies immediately to the large ecosystem of released checkpoints (including community fine-tunes) without distillation or trajectory redesign. The algebraic and optimality results (Theorems 5–6) are clean, the curvature-independent error decomposition (Theorem 9) and the straightness counter-example (Proposition 12) are useful conceptual contributions, and the empirical coverage across six model families with systematic ablations is unusually thorough for an inference-only method. The orthogonal composition with distillation (Z-Image-Turbo) is a practically valuable observation. These strengths make the paper a solid candidate for a methods-oriented ML venue.

major comments (3)
  1. Theorem 9 decomposes TJS error into E[∥et*∥²] + U(t*), yet the experiments never estimate either term separately. All reported curves are end-to-end quality metrics. Without a direct measurement (e.g., proxy residual of the velocity field, or Monte-Carlo estimates of conditional variance on synthetic Gaussian data), it remains unclear how much of the observed early-exit degradation is model error versus irreducible uncertainty. This is the load-bearing empirical precondition for Corollary 10 and for aggressive γ choices; a short diagnostic experiment would substantially strengthen the theory–experiment link.
  2. The main baselines are full-K ODE trajectories. Theorem 11 compares TJS to coarse Euler at matched NFE, but the practical competitor is a modern few-step solver (DPM-Solver++, UniPC, etc.) run for the same reduced NFE budget without an endpoint jump. Table 6 shows sampler-agnosticity at fixed k*=12, yet does not report quality of those solvers when stopped at the same NFE without decoding. A head-to-head at matched NFE (TJS-γ vs. solver with γK steps) is needed to isolate the gain attributable to the jump itself rather than simply using fewer steps.
  3. Practical selection of t* (or γ) is left as a user-chosen operating point. Tables 2 and 10 and the Pareto analysis show that different metrics saturate at different depths (CLIP early, ImageReward late), and complex prompts are noted as failure modes in §B.5. Without a cheap adaptive rule or a recommended default schedule (beyond the heuristic γ≈0.5), practitioners cannot reliably deploy TJS on new checkpoints or prompt distributions. A short adaptive procedure or a validated default table would make the method more actionable.
minor comments (6)
  1. Figure 1 caption and the surrounding text use both t and discrete step indices; a single consistent time convention (forward t:0 o1 or reverse) stated once in Preliminaries would reduce reader friction.
  2. In Algorithm 7 the notation mixes γ and t*; the main text later uses both γ and k*. Aligning on one primary symbol (with the other defined) would help.
  3. The claim that TJS can beat the full ODE (TJS-best overshoots in Table 1 and Z-Image-Turbo HPSv2/AES) is interesting; a one-sentence pointer in the main text to the discretization-error explanation already present in §B.5 would prevent misreading these as free quality gains.
  4. Related Work cites DPM-Solver with a broken reference marker (“?”). Please fix the bibliography entry.
  5. Supplementary §A.9 on DDIM is thorough; a one-paragraph condensation of the five numbered differences could usefully appear in the main Related Work or Discussion for readers who skip the supplement.
  6. Typos / spacing: “trainingcostthroughretraining”, “changestothetrajectory”, and similar missing spaces appear in the Introduction; a pass for word-boundary errors would improve polish.

Circularity Check

0 steps flagged

No significant circularity: decoder and MMSE optimality follow by linear algebra and standard ℓ2 Bayes properties from the affine-path definition; TJS is an inference policy, not a fitted prediction.

full rationale

The load-bearing chain is Definition 1 (affine path xt=αt x0+σt ε) → Lemma 2 (ut) → invertible linear system when Δt eq0 → closed-form decoder of Theorem 5. Theorem 6 then substitutes the Bayes-optimal velocity vigstar=E[ut|xt] and recovers E[x0|xt] by linearity of conditional expectation; the cross-term vanishing in Theorem 9 is the usual orthogonality of the MMSE residual. None of these steps introduce free parameters fitted to the reported NFE savings, nor do they rely on self-citation of a uniqueness theorem or an ansatz. γ (or t*) is an explicit user-chosen operating point, not a quantity whose value is forced by a prior fit. The comparison to DDIM is acknowledged as algebraic similarity with a different inference strategy; the paper does not rename a known empirical pattern as a new derivation. Experiments are external validation on off-the-shelf checkpoints. The derivation is therefore self-contained against its own inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 2 invented entities

The central claim rests on the standard definition of affine probability paths (already ubiquitous in diffusion/flow matching), the usual ℓ2 training objective, and the algebraic invertibility of the 2×2 path matrix when Δt ≠ 0. No new physical constants or free parameters are introduced; the early-exit time t* is a free user choice, not a fitted constant required for the theory. The only invented notions are the named property “endpoint decodability” and the algorithm TJS, both of which are definitional packaging of the algebra rather than new ontological entities.

free parameters (1)
  • early-exit fraction γ (or t*)
    User-chosen operating point that trades quality for speed; not fitted to force the theoretical claims, but the reported NFE savings depend on the chosen γ.
axioms (3)
  • domain assumption The generative process follows an affine probability path xt = αt x0 + σt ε with C1 schedules satisfying the usual boundary conditions.
    Stated as Definition 1; covers all standard VP/VE/EDM/linear-FM schedules used by the community.
  • domain assumption Models are trained under the standard ℓ2 (flow-matching or noise-prediction) objective, so the Bayes-optimal predictor is the conditional expectation.
    Used in Theorem 6 to identify the decoder with E[x0|xt].
  • standard math The path determinant Δt ≠ 0 for all t in (0,1] on the schedules of interest.
    Verified by direct differentiation for linear FM, VP, and VE/EDM in the supplement; fails only for pathological proportional decay.
invented entities (2)
  • endpoint decodability independent evidence
    purpose: Name the algebraic property that (xt, ut) uniquely determines x0 whenever Δt ≠ 0.
    Definitional packaging of the linear-system invertibility already present in the path definition; no new ontology.
  • Truncated Jump Sampling (TJS) independent evidence
    purpose: Operationalize early-exit decoding as a concrete inference algorithm.
    Algorithmic wrapper around the decoder; falsifiable by running the procedure on public checkpoints.

pith-pipeline@v1.1.0-grok45 · 52120 in / 2864 out tokens · 26567 ms · 2026-07-14T16:06:31.606560+00:00 · methodology

0 comments
read the original abstract

Diffusion and flow matching models generate high-quality samples, but their ODE samplers often need tens to hundreds of neural function evaluations (NFEs). This remains a practical challenge for released checkpoints, since many accelerators require additional design choices and training cost through retraining, distillation, or trajectory redesign. We investigate a different route based on $x$-prediction. During sampling, standard affine probability paths already expose $x_0$ information: an intermediate state and its path velocity determine a principled estimate of the clean sample. We formalize this property as \textbf{endpoint decodability} and show that the decoder is the minimum-MSE estimator $\mathbb{E}[x_0\mid x_t]$ under the usual $\ell_2$ objective. This yields \textbf{Truncated Jump Sampling} (TJS): stop the ODE at an early-exit time $t^*$ and return the decoded $x_0$. TJS requires no retraining, distillation, or architecture change. Across SDXL, SD3.5M, Z-Image-Turbo, and three class-conditional benchmarks, it reduces NFEs by 20--70\% with near-matched quality. The analysis also shows why endpoint prediction can work without straightening the trajectory, providing inference acceleration without trajectory redesign.

Figures

Figures reproduced from arXiv: 2607.06114 by Ang Gao, Xin Peng.

Figure 1
Figure 1. Figure 1: Endpoint decodability in action. Top: xt decoded directly (noisy at early steps). Bottom: x0 via endpoint decoding (clean at any step). strictly beat coarse Euler (Theorem 11). All standard pa￾rameterizations are equivalent at optimality (see Supple￾mentary Material, §A). Critically, straight trajectories are sufficient but not necessary (Proposition 12), challenging the foundation of Rectified Flow and Co… view at source ↗
Figure 2
Figure 2. Figure 2: Visual x0 predictions for CIFAR-10 (left), MNIST (center), and ImageNet-256 (right). MNIST saturate at k ∗≈16 (43% NFE saving); CIFAR-10/ImageNet-256 at k ∗≈26 (33%). 0 20 40 NFE 0 25 50 75 100 125 150 175 200 FID ↓ TJS best: 12.09 | Full ODE: 13.30 CIFAR-10 0 10 20 30 NFE 0 25 50 75 100 125 150 175 200 FID ↓ TJS best: 1.65 | Full ODE: 3.20 MNIST 0 20 40 NFE 0 50 100 150 200 250 300 350 FID ↓ TJS best: 15.… view at source ↗
Figure 3
Figure 3. Figure 3: FID vs. NFE for TJS on MNIST (30-step) and CIFAR-10/ImageNet-256 (40-step, CFG=1.0). [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual x0 predictions for SDXL (left) and SD3.5M (right). Saturation at k ∗≈19 (∼33% NFE saving). Theorem 11 (TJS–Euler Comparison). Let the affine path have C 2 coefficients. Compare two strategies at the same NFE budget N + 1: (a) Coarse Euler from t=0 to t=1 with step h = 1/N; (b) TJS-N, integrating to t ∗ = Nh via Euler then applying endpoint decoding. Assume E[∥et∥ 2 ] ≤ ε uniformly, and define Cα,σ =… view at source ↗
Figure 5
Figure 5. Figure 5: Speed vs. quality trade-off. Left: ImageReward against NFE, with 90% of full ODE quality marked per model. Right: [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Four-panel detailed view of T2I quality metrics (PickScore, ImageReward, HPSv2, CLIP) for SDXL and SD3.5M [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comprehensive multi-benchmark TJS convergence analysis. Five metric panels, each with six curves (3 benchmarks [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sampler ablation (SDXL, k ∗=12, DrawBench). Five common ODE solvers (DDIM, DPM++, LMS, PNDM, UniPC) are evaluated with identical k ∗ and K. Bar chart displays PickScore, CLIP, and ImageReward for each solver. The near-identical bar heights across all five solvers confirm that endpoint decoding quality is sampler-agnostic: the endpoint predictor extracts the same information regardless of the specific integ… view at source ↗
Figure 9
Figure 9. Figure 9: Schedule ablation (SDXL, DrawBench). Four noise schedules (Beta, Exponential, Karras, Laplace) evaluated across [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Step-count ablation (SDXL, DrawBench). ImageReward evaluated at three fixed integration fractions ( [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: CFG scale ablation (SDXL, DrawBench). ImageReward evaluated at CFG scales [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: CFG scale ablation on ImageNet-256 (class-conditional generation). Full TJS FID sweep (40 steps) at three CFG [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Visual x0 predictions for ImageNet-256 at CFG=1.25 (left) and CFG=1.5 (right). As k ∗ increases, image quality improves monotonically: global structure emerges first (k ∗=0–12), followed by texture detail (k ∗=18–26). The TJS-best (⋆) predictions at k ∗=26–32 are visually indistinguishable from or superior to the full 40-step ODE. Compare with [PITH_FULL_IMAGE:figures/full_fig_p028_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comprehensive TJS convergence analysis for Z-Image-Turbo ( [PITH_FULL_IMAGE:figures/full_fig_p029_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Visual x0 predictions for Z-Image-Turbo (K=10) at increasing k ∗ ∈ {0, 1, 2, 3, 4, 6, 8}, plus the full 10-step ODE as reference. Each row shows a different prompt. The visual progression confirms the quantitative findings: k ∗=0 (1 NFE) already produces semantically recognizable content; k ∗=2 (3 NFE) resolves fine details such as text rendering, facial features, and material textures; k ∗=3 (4 NFE) outp… view at source ↗
Figure 16
Figure 16. Figure 16: Direct visual comparison of xt (top row, the intermediate state at step k ∗ ) vs. x0 (bottom row, endpoint-decoded from the same xt) for Z-Image-Turbo (K=10). This figure provides the most direct illustration of Theorem 5: while xt remains corrupted by noise (top row, especially at early k ∗ ), the endpoint decoder recovers a clean, semantically coherent x0 (bottom row) from the exact same intermediate st… view at source ↗

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

87 extracted references · 10 linked inside Pith

  1. [1]

    Flow Matching for Generative Modeling , author=

  2. [2]

    Building Normalizing Flows with Stochastic Interpolants , author=

  3. [3]

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

  4. [4]

    Generative modeling by estimating gradients of the data distribution , author=

  5. [5]

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

  6. [6]

    Denoising diffusion probabilistic models , author=

  7. [7]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015. 2015

    2015

  8. [8]

    Denoising Diffusion Implicit Models , author=

  9. [9]

    2021 , organization=

    Improved denoising diffusion probabilistic models , author=. 2021 , organization=

    2021

  10. [10]

    arXiv preprint arXiv:2207.12598 , year=

    Classifier-Free Diffusion Guidance , author=. arXiv preprint arXiv:2207.12598 , year=

    Pith/arXiv arXiv

  11. [11]

    Diffusion models beat gans on image synthesis , author=

  12. [12]

    Progressive Distillation for Fast Sampling of Diffusion Models , author=

  13. [13]

    Masked Generative Distillation , author=

  14. [14]

    Consistency Models , author=

  15. [15]

    2023 , eprint=

    Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference , author=. 2023 , eprint=

    2023

  16. [16]

    Zero-shot text-to-image generation , author=

  17. [17]

    arXiv preprint arXiv:2204.06125 , year=

    Hierarchical text-conditional image generation with clip latents , author=. arXiv preprint arXiv:2204.06125 , year=

    Pith/arXiv arXiv

  18. [18]

    2022 , organization=

    GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models , author=. 2022 , organization=

    2022

  19. [19]

    Photorealistic text-to-image diffusion models with deep language understanding , author=

  20. [20]

    Transactions on Machine Learning Research , issn=

    Scaling Autoregressive Models for Content-Rich Text-to-Image Generation , author=. Transactions on Machine Learning Research , issn=. 2022 , url=

    2022

  21. [21]

    High-resolution image synthesis with latent diffusion models , author=

  22. [22]

    arXiv preprint arXiv:2307.01952 , year=

    Sdxl: Improving latent diffusion models for high-resolution image synthesis , author=. arXiv preprint arXiv:2307.01952 , year=

    Pith/arXiv arXiv

  23. [23]

    Scalable diffusion models with transformers , author=

  24. [24]

    Scaling rectified flow transformers for high-resolution image synthesis , author=

  25. [25]

    2026 , url=

    Improving Classifier-Free Guidance in Masked Diffusion: Low-Dim Theoretical Insights with High-Dim Impact , author=. 2026 , url=

    2026

  26. [26]

    Black-Forest-Labs. FLUX.1. 2024

    2024

  27. [27]

    arXiv preprint arXiv:2405.12230 , year=

    Qwen-Image Technical Report , author=. arXiv preprint arXiv:2405.12230 , year=

    Pith/arXiv arXiv

  28. [28]

    arXiv preprint arXiv:2502.17332 , year=

    Wan: Open and Advanced Large-Scale Video Generative Models , author=. arXiv preprint arXiv:2502.17332 , year=

    Pith/arXiv arXiv

  29. [29]

    PriorGrad: Improving Conditional Denoising Diffusion Models with Data-Dependent Adaptive Prior , author=

  30. [30]

    2021 , url=

    Diffusion Schr\"odinger Bridge with Applications to Score-Based Generative Modeling , author=. 2021 , url=

    2021

  31. [31]

    2025 , eprint=

    CFG-Zero*: Improved Classifier-Free Guidance for Flow Matching Models , author=. 2025 , eprint=

    2025

  32. [32]

    2024 , eprint=

    CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models , author=. 2024 , eprint=

    2024

  33. [33]

    Rectified

    Shreshth Saini and Shashank Gupta and Alan Bovik , booktitle=NeurIPS, year=. Rectified

  34. [34]

    2026 , eprint=

    CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance , author=. 2026 , eprint=

    2026

  35. [35]

    2026 , eprint=

    C ^2 FG: Control Classifier-Free Guidance via Score Discrepancy Analysis , author=. 2026 , eprint=

    2026

  36. [36]

    2025 , eprint=

    Warm Starts Accelerate Conditional Diffusion , author=. 2025 , eprint=

    2025

  37. [37]

    Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution , author=

  38. [38]

    Chen Chen and Pengsheng Guo and Liangchen Song and Jiasen Lu and Rui Qian and Tsu-Jui Fu and Xinze Wang and Wei Liu and Yinfei Yang and Alex Schwing , booktitle=NeurIPS, year=

  39. [39]

    Neural ordinary differential equations , author=

  40. [40]

    What uncertainties do we need in bayesian deep learning for computer vision? , author=

  41. [41]

    and Bottou, L

    Lecun, Y. and Bottou, L. and Bengio, Y. and Haffner, P. , journal=. Gradient-based learning applied to document recognition , year=

  42. [42]

    Learning multiple layers of features from tiny images , year =

    Krizhevsky, Alex and Hinton, Geoffrey , address =. Learning multiple layers of features from tiny images , year =

  43. [43]

    ImageNet: A large-scale hierarchical image database , year=

    Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Kai Li and Li Fei-Fei , booktitle=. ImageNet: A large-scale hierarchical image database , year=

  44. [44]

    2017 , isbn =

    Heusel, Martin and Ramsauer, Hubert and Unterthiner, Thomas and Nessler, Bernhard and Hochreiter, Sepp , title =. 2017 , isbn =

    2017

  45. [45]

    2016 , isbn =

    Salimans, Tim and Goodfellow, Ian and Zaremba, Wojciech and Cheung, Vicki and Radford, Alec and Chen, Xi , title =. 2016 , isbn =

    2016

  46. [46]

    2025 , eprint=

    SD3.5-Flash: Distribution-Guided Distillation of Generative Flows , author=. 2025 , eprint=

    2025

  47. [47]

    2016 , volume=

    Szegedy, Christian and Vanhoucke, Vincent and Ioffe, Sergey and Shlens, Jon and Wojna, Zbigniew , booktitle=CVPR, title=. 2016 , volume=

    2016

  48. [48]

    arXiv , primaryClass=

    Flow Matching in Latent Space , author=. arXiv , primaryClass=. 2023 , eprint=

    2023

  49. [49]

    2020 , eprint=

    An Ensemble of Simple Convolutional Neural Network Models for MNIST Digit Recognition , author=. 2020 , eprint=

    2020

  50. [50]

    Deep Residual Learning for Image Recognition , author=

  51. [51]

    generation: Taming optimization dilemma in latent diffusion models , author=

    Reconstruction vs. generation: Taming optimization dilemma in latent diffusion models , author=

  52. [52]

    2026 IEEE International Conference on Robotics and Automation (ICRA) , year=

    Dense-Jump Flow Matching with Non-Uniform Time Scheduling for Robotic Policies: Mitigating Multi-Step Inference Degradation , author=. 2026 IEEE International Conference on Robotics and Automation (ICRA) , year=

    2026

  53. [53]

    Elucidating the Design Space of Diffusion-Based Generative Models , author=

  54. [54]

    InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation , author=

  55. [55]

    ICML 2023 Workshop on Structured Probabilistic Inference and Generative Modeling , year=

    BOOT: Data-free Distillation of Denoising Diffusion Models with Bootstrapping , author=. ICML 2023 Workshop on Structured Probabilistic Inference and Generative Modeling , year=

    2023

  56. [56]

    Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation , author=

  57. [57]

    arXiv preprint arXiv:2306.09341 , year=

    Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis , author=. arXiv preprint arXiv:2306.09341 , year=

    Pith/arXiv arXiv

  58. [58]

    ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation , author=

  59. [59]

    Journal of the American Statistical Association , volume =

    Bradley Efron , title =. Journal of the American Statistical Association , volume =. 2011 , publisher =

    2011

  60. [60]

    IEEE Transactions on Information Theory , volume=

    Mutual information and minimum mean-square error in Gaussian channels , author=. IEEE Transactions on Information Theory , volume=. 2005 , publisher=

    2005

  61. [61]

    Learning Transferable Visual Models From Natural Language Supervision , author=

  62. [62]

    2025 , issn =

    DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models , journal =. 2025 , issn =. doi:10.1007/s11633-025-1562-4 , url =

    doi:10.1007/s11633-025-1562-4 2025

  63. [63]

    SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations , author=

  64. [64]

    Variational Diffusion Models , author=

  65. [65]

    Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models , author=

  66. [66]

    Fast Sampling of Diffusion Models via Operator Learning , author =

  67. [67]

    Pseudo Numerical Methods for Diffusion Models on Manifolds , author=

  68. [68]

    Trajectory Consistency Distillation: Improved Few-Step Image Generation , author=

  69. [69]

    Flow Matching for Conditional Generation with Optimal Transport , author=

  70. [70]

    Improved Techniques for Training GANs , author=

  71. [71]

    Flow Matching on Manifolds , author=

  72. [72]

    2024 , eprint=

    Improving and generalizing flow-based generative models with minibatch optimal transport , author=. 2024 , eprint=

    2024

  73. [73]

    PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores , author=

  74. [74]

    TADA: Improved Diffusion Sampling with Training-Free Augmented Dynamics , author=

  75. [75]

    arXiv preprint arXiv:2505.24210 , year=

    STORK: Faster Diffusion and Flow Matching Sampling by Resolving Both Stiffness and Structure-Dependence , author=. arXiv preprint arXiv:2505.24210 , year=

    arXiv

  76. [76]

    SADA: Stability-Guided Adaptive Diffusion Acceleration , author=

  77. [77]

    arXiv preprint arXiv:2502.11420 , year=

    TreeG: Tree Search-Based Path Steering Guidance for Diffusion and Flow Matching , author=. arXiv preprint arXiv:2502.11420 , year=

    Pith/arXiv arXiv

  78. [78]

    A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models , author=

  79. [79]

    DuoDiff: Accelerating Diffusion Models with a Dual-Backbone Approach , author=

  80. [80]

    The Diffusion Duality , author=

Showing first 80 references.