Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Two-step diffusion models can be aligned to arbitrary rewards via a learned differentiable latent-space surrogate reward.

desk verdict Useful new method for aligning two-step diffusion models via latent surrogate rewards; the empirical case is strong, but single-run evaluation and best-FID checkpoint selection need addressing. read the letter →

arxiv 2411.15247 v3 pith:ZS3O5LMH submitted 2024-11-22 cs.LG

Zhiwei Jia , Yuesong Nan , Huixi Zhao , Gengdai Liu This is my paper · ORCID
classification cs.LG
keywords rewardfine-tuningdiffusionmodelslatentconsistencyfew-stepimagegenerationsurrogateoff-policyexplorationtext-to-imagealignmentreinforcementlearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that existing reinforcement-learning fine-tuning methods fail for one- and two-step distilled diffusion models because the sampling map is nearly deterministic and highly non-smooth, so policy-gradient objectives degenerate and gradient estimation is unstable. In response, LaSRO learns a differentiable surrogate reward in the latent space of SDXL that converts arbitrary rewards, including non-differentiable ones, into gradient guidance for the distilled generator. Training uses Bradley-Terry preference pairs ranked by the true reward on outputs of the base model, with off-policy exploration by resampling initial noise, and an online adaptation stage that relabels new samples. The paper reports consistent gains in Image Reward, attribute binding, and text alignment for latent consistency models, with better results than DDPO, Diffusion-DPO, and GORS-LCM baselines.

What carries the argument

The load-bearing object is the latent-space surrogate reward $R_\psi(z,c)$: a small CNN prediction head mounted on the UNet encoder of a pre-trained latent diffusion model, trained with a Bradley-Terry binary cross-entropy loss on winning/losing latent pairs selected by the true reward. During fine-tuning, LaSRO minimizes a normalized, clipped version of this surrogate score at both the first and second sampling steps, added to the original latent consistency distillation loss as a regularizer. The online adaptation substage refits the surrogate on newly generated pairs from a replay buffer, which is intended to keep the surrogate aligned with the true reward as the generator drifts.

What would settle it

On a fixed held-out set of prompts, record the true reward and the surrogate's predicted score throughout LaSRO fine-tuning of a two-step LCM: if the surrogate score keeps increasing while the true reward plateaus or declines, or if FID worsens while Image Reward improves, the surrogate has been hacked and the central claim fails in that regime. A second check is to ablate online adaptation and compare true-reward curves; if removing it does not reduce the true reward, the surrogate's adaptation is not what carries the result.

Watch

Extended reading notes

Core claim

The central claim is that fine-tuning ultra-fast (one- or two-step) distilled diffusion models with arbitrary rewards should be done by learning a differentiable surrogate reward in latent space rather than by estimating policy gradients or performing reward-weighted regression. The surrogate, trained with a Bradley-Terry pairwise loss on samples from the pre-fine-tuning generator and ranked by the target reward, provides gradient signal through both sampling steps and makes exploration possible through resampling of initial noise. The paper argues that this is functionally a value-based RL approach: for a horizon-two MDP with discount zero, the temporal-difference regression target reduces to predicting the true reward from the latent, and off-policy exploration explains the method's stability. On this basis the paper reports that LaSRO outperforms RLCM/DDPO, PSO/Diffusion-DPO, and GORS-LCM across three reward objectives and on both in-distribution and out-of-distribution prompts.

Load-bearing premise

The load-bearing premise is that the surrogate reward, trained on pairs drawn from the pre-fine-tuning model, continues to provide gradient directions that align with the true reward after the generator drifts off that training distribution; online relabeling updates the surrogate's values but does not by itself guarantee that its gradients cannot be exploited.

Editorial extensions

If this is right

  • Arbitrary black-box rewards, such as scores built from captioning models, visual question answering, or other non-differentiable pipelines, can be used to fine-tune one- and two-step image generators.
  • The diagnosed failure modes (degenerated objectives, non-smooth latent mappings, and insufficient stochasticity) give a concrete explanation for why policy-gradient and DPO-style methods are unstable on step-distilled models.
  • LaSRO's reported results imply that a single pre-trained latent diffusion encoder can serve as a reusable reward backbone across different reward objectives, including out-of-distribution prompt sets.
  • The value-based connection suggests that short-horizon generation MDPs are better handled by learned value or reward functions with off-policy exploration than by on-policy policy-gradient estimation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, a direct way to test the surrogate for reward hacking is to track surrogate score and true reward on a fixed held-out prompt set: if the surrogate rises while the true reward plateaus or drops, the gradient signal has been exploited.
  • Beyond the paper, the value-based interpretation suggests that replacing the Bradley-Terry head with a temporal-difference Q-estimate over latent states could extend the same scheme to longer sampling horizons or sequential latent samplers.
  • Beyond the paper, because exploration comes from resampling initial and injected noise rather than from policy stochasticity, a LaSRO-style surrogate could plausibly fine-tune fully deterministic one-step generators, which policy-gradient methods cannot address.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper addresses reward fine-tuning of step-distilled diffusion models that generate images in at most two steps. It argues that policy-gradient RL methods such as DDPO and reward-weighted-regression methods such as Diffusion-DPO are ineffective for these models because the two-step sampler is nearly deterministic, its mapping from noisy to clean latents is non-smooth, and its stochasticity is too limited for exploration. The proposed method, LaSRO, learns a differentiable surrogate reward Rψ in the latent space of SDXL, using a pre-trained UNet encoder plus a small prediction head, trained by a Bradley-Terry pairwise loss on samples generated by the baseline LCM. Fine-tuning then minimizes a weighted combination of the original LCM distillation loss and normalized, clipped surrogate-reward losses for the first and second sampling steps, with periodic online adaptation of the surrogate. Experiments on Image Reward, an attribute-binding score, and a text-alignment score report large improvements over the LCM baseline and over RLCM/DDPO, PSO/Diffusion-DPO, and GORS-LCM, and ablations support several design choices. The paper also sketches a connection between LaSRO and value-based RL.

Significance. If the reported results are reproducible, LaSRO fills a genuine practical gap: aligning ultra-fast distilled generators to arbitrary, possibly non-differentiable rewards without relying on high-variance policy-gradient estimates. The method is simple, builds on off-the-shelf components, and the experimental section covers three distinct reward signals plus ablations on backbone choice, loss terms, and online adaptation. The authors also provide a plausible theoretical interpretation via value-based RL, though it is informal. The main weaknesses are that the empirical evidence is single-run and summarized by best-FID checkpoint selection, and that the surrogate reward's alignment with the true reward under distribution shift is not directly measured. These issues are addressable and do not invalidate the core idea, but they currently limit the strength of the central claim.

major comments (3)
  1. [Sec. 6.1 and App. E; Table 1] The summary numbers in Table 1 are obtained by selecting, for each run, the checkpoint with the lowest FID on MJHQ-30K (described in App. E), and the curves in Fig. 4 appear to come from a single run with no error bars. Because the selection criterion is itself an evaluation metric, the headline comparison can reflect favorable checkpoint selection rather than stable improvement under a pre-specified training budget. Please report mean ± standard deviation over at least three seeds at fixed update counts for all compared methods, and present the full FID-versus-Image-Reward curves of Fig. 8 as primary evidence for the tradeoff behavior.
  2. [Sec. 5.1, Eq. (10); Alg. 2] The surrogate reward Rψ is learned from pairs generated by the pre-fine-tuning LCM and is then used to provide gradient guidance after the LCM parameters have drifted. Online adaptation in Alg. 2 (lines 12–15) re-labels only a reset replay buffer and updates Rψ for N2 steps; nothing in the procedure constrains the local gradient field of Rψ to remain aligned with the true reward r on the distribution of latents generated by the fine-tuned model. Since Eq. (11) directly ascends S[Rψ], the method can in principle increase Rψ without improving r (reward hacking). Please add calibration evidence, for example rank correlation between Rψ and r on held-out prompts from several fine-tuned checkpoints, and a training curve showing the true reward alongside the surrogate reward. The 'No Online Adapt' ablation shows that adaptation helps, but it does not measure alignment.
  3. [Sec. 6.3 and App. C] The claimed connection to value-based RL is stated as an equivalence, but Eq. (16) replaces the actual pairwise logistic loss in Eq. (10) with an L2 regression to the true reward, and the argument relies on γ=0 and horizon H=2. If this is intended as a formal theorem, a proof is needed; otherwise the text should explicitly label the connection as an analogy. As written, the 'theoretical insight' wording overstates what is established and could mislead readers about the guarantees of the method.
minor comments (5)
  1. [Sec. 4.2, Eq. (9)] Equation (9) appears to define J_deg as an expression containing ∇θ log pθ, which is a gradient, not an objective. Please either define J_deg as the objective and then give its gradient separately, or label the displayed quantity as ∇θJ_deg.
  2. [Sec. 5.3 and App. C, Eq. (16)] The text says the TD loss is 'essentially the surrogate reward loss Lsurr in an L2 regression form', but Lsurr in Eq. (10) is a pairwise classification loss. Please clarify that this is an informal analogy rather than a mathematical equivalence.
  3. [Sec. 5.1] The sentence 'the winning and losing pair (zw, zl) are selected' has a subject-verb agreement error; it should be 'is selected' or 'pairs are selected'. The phrase 'more on this shortly' is also informal for a journal.
  4. [Sec. 6.3] The claim that LaSRO is 'relatively robust against manual noise perturbation of the GT. reward' is made without showing the experiment or the perturbation protocol. Please either provide the details and results or remove the claim.
  5. [General] No code or dataset release is mentioned. Given the reliance on a custom LCM trained from SSD-1B and a new 873k-prompt set, a reproducibility statement or release commitment would be valuable.

Circularity Check

0 steps flagged · score 1.0 of 10

No load-bearing circularity: LaSRO fits a surrogate reward to the target reward and evaluates on the same target reward, which is standard reward-model optimization rather than a derivation that reduces to its own inputs.

full rationale

The paper's central empirical claim is that LaSRO improves two-step LCMs on external reward metrics. The surrogate Rψ is fitted to the target reward r via the Bradley–Terry loss (Eq. 10), and the fine-tuning loss (Eq. 11) maximizes the normalized surrogate. This is a fitted predictor used for optimization, not a circular derivation: the improvement is verified with the true reward r on held-out prompts (T2I-prompt-873K test split and MJHQ-30K) and against baselines, and the ablation 'w\ GT.gradient' compares LaSRO against direct true-reward gradients. The value-based RL connection (Sec. 5.3 and App. C) is explicitly presented as a theoretical insight and is not used to derive the empirical results. The only self-citations ([27], [28]) support a generic 'sharpness' remark and are not load-bearing. The FID-based checkpoint selection described in App. E is a benchmark-validity concern, not a circular step, because no parameter of the surrogate or the LCM is defined in terms of the reported MJHQ-30K Image Reward score.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

LaSRO is an empirical deep learning pipeline. It introduces a trained surrogate reward Rψ with several hand-chosen hyperparameters (loss coefficients, sample counts, learning rates, EMA rate, clipping percentile) and assumes that the two-step LCM mapping is non-smooth, that the Bradley-Terry surrogate generalizes to the fine-tuning distribution, and that consistency-loss regularization prevents reward overoptimization. No new physical entities, forces, particles, dimensions, or formal axioms are introduced.

free parameters (6)
  • Loss coefficients (c, c1, c2) in Eq. 11 = c=500, c1=0.5, c2=1.0
    Chosen by hand to balance distillation regularization, first-step reward, and second-step reward. Reported in Appendix D.
  • Ns, samples per prompt for winner/loser pair selection = 1 for Image Reward, 6 otherwise
    Tuned to reward discreteness and exploration difficulty. Reported in Appendix D.
  • Online adaptation schedule (N1, N2) = (1,1) for Image Reward, (10,20) for Attribute Binding, (10,50) for Text Alignment
    Manually set per task; controls the alternation between fine-tuning the diffusion model and adapting the surrogate reward. Reported in Appendix D.
  • Learning rates eta1 and eta2 = 3e-6 and 3e-8
    Chosen by hand for all three rewards. Reported in Appendix D.
  • EMA rate mu for checkpoint reporting = 0.95
    Used to stabilize reported results across iterations. Reported in Appendix D.
  • Moving 90th percentile R90% in normalization function S = computed online
    Design choice for clipping normalized surrogate rewards, described in Appendix B and used in Eq. 12 and Eq. 13.
assumptions (6)
  • domain assumption The two-step LCM mapping fθ has high local Lipschitz constant that grows with input noise level, making policy-gradient and reward-weighted regression methods ineffective.
    Empirically estimated in Appendix A with N=1000 pairs (Fig. 7), not proven. It grounds the motivation in Section 4.3.
  • domain assumption A Bradley-Terry-trained latent surrogate Rψ provides useful gradient signal for optimizing the true reward r on unseen prompts and on the shifted fine-tuning distribution.
    Assumed throughout Section 5; online adaptation partially addresses distribution shift, but generalization and resistance to reward hacking are not guaranteed.
  • domain assumption The consistency-distillation loss Llcm acts as sufficient regularization against reward overoptimization during fine-tuning.
    Used in Eq. 11 and discussed in Appendix E; there is no formal guarantee that this regularization prevents reward hacking.
  • domain assumption The SDXL UNet encoder as backbone for Rψ is more effective than CLIP or BLIP and operates in a space where reward-relevant image details are preserved.
    Validated only through the ablations in Fig. 5 (w/ CLIP, w/ BLIP), not derived from first principles.
  • standard math The Bradley-Terry model as a pairwise binary cross-entropy loss is a valid objective for learning a reward surrogate.
    Standard statistical model for paired comparisons, used in Eq. 10.
  • domain assumption Pre-trained VAE and LCM components used for training and evaluation behave reliably under the transformations in the pipeline.
    The paper relies on SDXL VAE, SSD-1B teacher, and LCM student without re-validating their properties.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward." pith.science (2026). https://pith.science/paper/ZS3O5LMH

@misc{pith2026241115247,
  author       = {Pith},
  title        = {Pith review of: Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZS3O5LMH}},
  note         = {Machine review of arXiv:2411.15247}
}
abstract

Recent research has shown that fine-tuning diffusion models (DMs) with arbitrary rewards, including non-differentiable ones, is feasible with reinforcement learning (RL) techniques, enabling flexible model alignment. However, applying existing RL methods to step-distilled DMs is challenging for ultra-fast ($\le2$-step) image generation. Our analysis suggests several limitations of policy-based RL methods such as PPO or DPO toward this goal. Based on the insights, we propose fine-tuning DMs with learned differentiable surrogate rewards. Our method, named LaSRO, learns surrogate reward models in the latent space of SDXL to convert arbitrary rewards into differentiable ones for effective reward gradient guidance. LaSRO leverages pre-trained latent DMs for reward modeling and tailors reward optimization for $\le2$-step image generation with efficient off-policy exploration. LaSRO is effective and stable for improving ultra-fast image generation with different reward objectives, outperforming popular RL methods including DDPO and Diffusion-DPO. We further show LaSRO's connection to value-based RL, providing theoretical insights. See our webpage \href{https://sites.google.com/view/lasro}{here}.

Figures

Figures reproduced from arXiv: 2411.15247 by the authors.

Figure 2
Figure 2. (left) Given astronaut riding a horse as c, fixed ini￾tial noise xτ0 and guidance scale, the reduction of noise-injecting steps leads to less diverse images generated from SSD-1B [17] and its LCM, and thus harder exploration for pθ(xτH |xτ0 , c). (right) Illustration of the LCM mapping fθ. We show the growing empir￾ical local Lipchitz of the LCM mapping from a noisy image to a clean image and to a generic image qual… view at source ↗
Figure 3
Figure 3. Training pipeline of the fine-tuning stage of LaSRO. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. (left) LaSRO vs. GORS-LCM (other baselines fail) for fine-tuning LCMs with Attribute Binding Score and Text Align￾ment Score (all two-step perf. results). (right) Results from abla￾tion studies that validate LaSRO’s design (see Sec. 6.3). # steps resolution Image Reward SDXL-Turbo [61] 1 5122 0.817 SDXL-Turbo 2 5122 0.839 + LaSRO (ours) 1 5122 0.904 + LaSRO (ours) 2 5122 0.957 SDXL-Lightning [38] 2 10242 0.634 SSD-1… view at source ↗
Figures from the paper (3 more)
Figure 7
Figure 7. Figure 7: We show the growing empirical local Lipchitz of the mapping that underlies two-step LCM image generation. The x-axis is the [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: The curves that present the tradeoff between FID and Image Reward for the LCM baseline and the fine-tuned versions with all [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Additional visual comparison of two-step LCM results from the LCM baseline and the fine-tuned LCM via LaSRO with Image [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Selective Timestep Weighting and Advantage-Based Replay for Sample-Efficient Diffusion RLHF

    cs.LG 2026-07 conditional novelty 5.0 of 10

    Two plug-and-play strategies — per-timestep advantage weighting and advantage-based trajectory replay — improve diffusion RLHF sample efficiency up to 6× across five reward functions.

  2. VARD: Efficient and Dense Fine-Tuning for Diffusion Models with Value-based RL

    cs.CV 2025-05 conditional novelty 5.0 of 10

    VARD fine-tunes diffusion models by backpropagating through a learned value function that assigns dense, differentiable reward estimates to every intermediate denoising step, with KL regularization keeping the model n...

Reference graph

Works this paper leans on

92 extracted references · 29 canonical work pages · cited by 2 Pith papers

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 ,

    arXiv

  2. [2]

    Training diffusion models with reinforce- ment learning

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforce- ment learning. arXiv preprint arXiv:2305.13301, 2023. 1, 2, 3, 4, 6, 7, 8, 5

    arXiv 2023

  3. [3]

    Ralph Allan Bradley and Milton E. Terry. Rank analysis of incomplete block designs: I. the method of paired compar- isons. Biometrika, 39(3-4):324–345, 1952. 5

    1952

  4. [4]

    Find: Fine- tuning initial noise distribution with policy optimization for diffusion models

    Changgu Chen, Libing Yang, Xiaoyan Yang, Lianggangxu Chen, Gaoqi He, Changbo Wang, and Yang Li. Find: Fine- tuning initial noise distribution with policy optimization for diffusion models. In Proceedings of the 32nd ACM Interna- tional Conference on Multimedia , pages 6735–6744, 2024. 3

    2024

  5. [5]

    Diffusion policy: Visuomotor policy learning via action dif- fusion

    Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. Diffusion policy: Visuomotor policy learning via action dif- fusion. The International Journal of Robotics Research, page 02783649241273668, 2023. 2

    2023

  6. [6]

    Simple drop-in lora condition- ing on attention layers will improve your diffusion model

    Joo Young Choi, Jaesung R Park, Inkyu Park, Jaewoong Cho, Albert No, and Ernest K Ryu. Simple drop-in lora condition- ing on attention layers will improve your diffusion model. arXiv preprint arXiv:2405.03958, 2024. 4

    arXiv 2024

  7. [7]

    Directly fine-tuning diffusion models on differentiable re- wards

    Kevin Clark, Paul Vicol, Kevin Swersky, and David J Fleet. Directly fine-tuning diffusion models on differentiable re- wards. arXiv preprint arXiv:2309.17400, 2023. 3, 8

    arXiv 2023

  8. [8]

    Prdp: Proximal reward difference prediction for large-scale reward finetuning of diffusion models

    Fei Deng, Qifei Wang, Wei Wei, Tingbo Hou, and Matthias Grundmann. Prdp: Proximal reward difference prediction for large-scale reward finetuning of diffusion models. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7423–7433, 2024. 3, 5, 4

    2024

Show all 92 references
  1. [9]

    Raft: Reward ranked finetuning for generative foundation model alignment

    Hanze Dong, Wei Xiong, Deepanshu Goyal, Yihan Zhang, Winnie Chow, Rui Pan, Shizhe Diao, Jipeng Zhang, Kashun Shum, and Tong Zhang. Raft: Reward ranked finetuning for generative foundation model alignment. arXiv preprint arXiv:2304.06767, 2023. 3

    2023 arXiv

  2. [10]

    Sigmoid- weighted linear units for neural network function approxima- tion in reinforcement learning

    Stefan Elfwing, Eiji Uchibe, and Kenji Doya. Sigmoid- weighted linear units for neural network function approxima- tion in reinforcement learning. Neural networks, 107:3–11,

  3. [11]

    Optimizing ddpm sampling with shortcut fine-tuning

    Ying Fan and Kangwook Lee. Optimizing ddpm sampling with shortcut fine-tuning. arXiv preprint arXiv:2301.13362,

    arXiv

  4. [12]

    Dpok: Reinforcement learning for fine-tuning text-to-image diffu- sion models

    Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Moham- mad Ghavamzadeh, Kangwook Lee, and Kimin Lee. Dpok: Reinforcement learning for fine-tuning text-to-image diffu- sion models. Advances in Neural Information Processing Systems, 3...

    2023

  5. [13]

    Re- inforcement learning for fine-tuning text-to-image diffusion models

    Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Moham- mad Ghavamzadeh, Kangwook Lee, and Kimin Lee. Re- inforcement learning for fine-tuning text-to-image diffusion models. Advances in Neural Information Processing Sys- tems, 36, 2024. 3

    2024

  6. [14]

    Sharpness-aware minimization for efficiently improving generalization

    Pierre Foret, Ariel Kleiner, Hossein Mobahi, and Behnam Neyshabur. Sharpness-aware minimization for efficiently improving generalization. arXiv preprint arXiv:2010.01412,

    2010 arXiv

  7. [15]

    Scaling laws for reward model overoptimization

    Leo Gao, John Schulman, and Jacob Hilton. Scaling laws for reward model overoptimization. InInternational Conference on Machine Learning, pages 10835–10866. PMLR, 2023. 7

    2023

  8. [16]

    Leveraging reward gradients for reinforcement learning in differentiable physics simulations

    Sean Gillen and Katie Byl. Leveraging reward gradients for reinforcement learning in differentiable physics simulations. arXiv preprint arXiv:2203.02857, 2022. 5

    2022 arXiv

  9. [17]

    Jaddipal, Harish Prabhala, Sayak Paul, and Patrick V on Platen

    Yatharth Gupta, Vishnu V . Jaddipal, Harish Prabhala, Sayak Paul, and Patrick V on Platen. Progressive knowledge distil- lation of stable diffusion xl using layer level loss, 2024. 4, 6, 2, 3

    2024

  10. [18]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. Advances in neural information processing systems , 30, 2017. 7, 5

    2017

  11. [19]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 1, 2, 3

    2020

  12. [20]

    Gritsenko, Diederik P

    Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey A. Gritsenko, Diederik P. Kingma, Ben Poole, Mohammad Norouzi, David J. Fleet, and Tim Sali- mans. Imagen video: High definition video generation with diffusion models. ArXiv, abs/2210.02303, 2022. 2

    arXiv

  13. [21]

    Video dif- fusion models

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models. Advances in Neural Information Processing Systems, 35:8633–8646, 2022. 2

    2022

  14. [22]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 4

    2021 arXiv

  15. [23]

    Difftaichi: Differentiable programming for physical simulation

    Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, and Fr´edo Durand. Difftaichi: Differentiable programming for physical simulation. arXiv preprint arXiv:1910.00935, 2019. 5

    1910 arXiv

  16. [24]

    T2i-compbench: A comprehensive bench- mark for open-world compositional text-to-image genera- tion

    Kaiyi Huang, Kaiyue Sun, Enze Xie, Zhenguo Li, and Xihui Liu. T2i-compbench: A comprehensive bench- mark for open-world compositional text-to-image genera- tion. Advances in Neural Information Processing Systems , 36:78723–78747, 2023. 1, 3, 7, 4

    2023

  17. [25]

    Plasticinelab: A soft-body manipulation benchmark with differentiable physics

    Zhiao Huang, Yuanming Hu, Tao Du, Siyuan Zhou, Hao Su, Joshua B Tenenbaum, and Chuang Gan. Plasticinelab: A soft-body manipulation benchmark with differentiable physics. arXiv preprint arXiv:2104.03311, 2021. 5

    2021 arXiv

  18. [26]

    Planning with diffusion for flexible behavior synthe- sis

    Michael Janner, Yilun Du, Joshua B Tenenbaum, and Sergey Levine. Planning with diffusion for flexible behavior synthe- sis. arXiv preprint arXiv:2205.09991, 2022. 2 9

    2022 arXiv

  19. [27]

    Information-theoretic local minima characterization and regularization

    Zhiwei Jia and Hao Su. Information-theoretic local minima characterization and regularization. In International Confer- ence on Machine Learning, pages 4773–4783. PMLR, 2020. 5

    2020

  20. [28]

    Semantically robust unpaired image translation for data with unmatched seman- tics statistics

    Zhiwei Jia, Bodi Yuan, Kangkang Wang, Hong Wu, David Clifford, Zhiqiang Yuan, and Hao Su. Semantically robust unpaired image translation for data with unmatched seman- tics statistics. In Proceedings of the IEEE/CVF international conference on computer vision, pages 14273–1428...

    2021

  21. [29]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation

    Yuval Kirstain, Adam Polyak, Uriel Singer, Shahbuland Ma- tiana, Joe Penna, and Omer Levy. Pick-a-pic: An open dataset of user preferences for text-to-image generation. Advances in Neural Information Processing Systems , 36: 36652–36663, 2023. 1, 3, 5, 7

    2023

  22. [30]

    Actor-critic algorithms

    Vijay Konda and John Tsitsiklis. Actor-critic algorithms. Ad- vances in neural information processing systems , 12, 1999. 5

    1999

  23. [31]

    Diffwave: A versatile diffusion model for audio synthesis

    Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. Diffwave: A versatile diffusion model for audio synthesis. arXiv preprint arXiv:2009.09761, 2020. 2

    2009 arXiv

  24. [32]

    Plastic: Improving input and label plasticity for sample efficient reinforcement learning

    Hojoon Lee, Hanseul Cho, Hyunseung Kim, Daehoon Gwak, Joonkee Kim, Jaegul Choo, Se-Young Yun, and Chul- hee Yun. Plastic: Improving input and label plasticity for sample efficient reinforcement learning. Advances in Neural Information Processing Systems, 36, 2024. 5

    2024

  25. [33]

    Aligning text- to-image models using human feedback

    Kimin Lee, Hao Liu, Moonkyung Ryu, Olivia Watkins, Yuqing Du, Craig Boutilier, Pieter Abbeel, Mohammad Ghavamzadeh, and Shixiang Shane Gu. Aligning text- to-image models using human feedback. arXiv preprint arXiv:2302.12192, 2023. 3

    2023 arXiv

  26. [34]

    Playground v2.5: Three insights towards enhancing aesthetic quality in text-to-image genera- tion, 2024

    Daiqing Li, Aleks Kamko, Ehsan Akhgari, Ali Sabet, Lin- miao Xu, and Suhail Doshi. Playground v2.5: Three insights towards enhancing aesthetic quality in text-to-image genera- tion, 2024. 7

    2024

  27. [35]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Interna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 1, 3, 5, 7, 4

    2022

  28. [36]

    Reward guided latent consistency distillation

    Jiachen Li, Weixi Feng, Wenhu Chen, and William Yang Wang. Reward guided latent consistency distillation. arXiv preprint arXiv:2403.11027, 13, 2024. 3, 5, 8, 4

    2024 arXiv

  29. [37]

    Step-aware prefer- ence optimization: Aligning preference with denoising per- formance at each step

    Zhanhao Liang, Yuhui Yuan, Shuyang Gu, Bohan Chen, Tiankai Hang, Ji Li, and Liang Zheng. Step-aware prefer- ence optimization: Aligning preference with denoising per- formance at each step. arXiv preprint arXiv:2406.04314, 2 (5):7, 2024. 3

    2024 arXiv

  30. [38]

    Sdxl- lightning: Progressive adversarial diffusion distillation

    Shanchuan Lin, Anran Wang, and Xiao Yang. Sdxl- lightning: Progressive adversarial diffusion distillation. arXiv preprint arXiv:2402.13929, 2024. 4, 5, 6, 7, 3

    2024 arXiv

  31. [39]

    Flow matching for generative mod- eling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling. arXiv preprint arXiv:2210.02747, 2022. 1

    2022 arXiv

  32. [40]

    Audi- oldm: Text-to-audio generation with latent diffusion models

    Haohe Liu, Zehua Chen, Yi Yuan, Xinhao Mei, Xubo Liu, Danilo Mandic, Wenwu Wang, and Mark D Plumbley. Audi- oldm: Text-to-audio generation with latent diffusion models. arXiv preprint arXiv:2301.12503, 2023. 2

    2023 arXiv

  33. [41]

    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. 1

    2022 arXiv

  34. [42]

    Instaflow: One step is enough for high-quality diffusion- based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion- based text-to-image generation. In The Twelfth International Conference on Learning Representations, 2023. 1, 3

    2023

  35. [43]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in Neural Information Processing Systems , 35:5775–5787,

  36. [44]

    Llmscore: Unveiling the power of large language models in text-to-image synthesis evaluation

    Yujie Lu, Xianjun Yang, Xiujun Li, Xin Eric Wang, and William Yang Wang. Llmscore: Unveiling the power of large language models in text-to-image synthesis evaluation. Ad- vances in Neural Information Processing Systems, 36, 2024. 1, 3

    2024

  37. [45]

    Latent consistency models: Synthesizing high- resolution images with few-step inference

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high- resolution images with few-step inference. arXiv preprint arXiv:2310.04378, 2023. 2, 3, 6, 7

    2023 arXiv

  38. [46]

    Tuning timestep- distilled diffusion model using pairwise sample optimiza- tion

    Zichen Miao, Zhengyuan Yang, Kevin Lin, Ze Wang, Zicheng Liu, Lijuan Wang, and Qiang Qiu. Tuning timestep- distilled diffusion model using pairwise sample optimiza- tion. arXiv preprint arXiv:2410.03190 , 2024. 2, 3, 4, 5, 7

    2024 arXiv

  39. [47]

    Playing atari with deep reinforcement learning

    V olodymyr Mnih. Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013. 6

    2013 arXiv

  40. [48]

    Rl for consistency models: Faster reward guided text-to-image generation

    Owen Oertell, Jonathan D Chang, Yiyi Zhang, Kiant ´e Brant- ley, and Wen Sun. Rl for consistency models: Faster reward guided text-to-image generation. arXiv preprint arXiv:2404.03673, 2024. 2, 3, 7, 4

    2024 arXiv

  41. [49]

    Reinforcement learning by reward-weighted regression for operational space control

    Jan Peters and Stefan Schaal. Reinforcement learning by reward-weighted regression for operational space control. In Proceedings of the 24th international conference on Machine learning, pages 745–750, 2007. 3, 5

    2007

  42. [50]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 1, 3, 5

    2023 arXiv

  43. [51]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. ArXiv, abs/2209.14988, 2022. 2

    2022 arXiv

  44. [52]

    Aligning text-to-image diffusion models with reward backpropagation

    Mihir Prabhudesai, Anirudh Goyal, Deepak Pathak, and Katerina Fragkiadaki. Aligning text-to-image diffusion models with reward backpropagation. arXiv preprint arXiv:2310.03739, 2023. 3

    2023 arXiv

  45. [53]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  46. [54]

    Direct preference optimization: Your language model is secretly a 10 reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a 10 reward model. Advances in Neural Information Processing Systems, 36, 2024. 1, 3, 5

    2024

  47. [55]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In International confer- ence on machine learning, pages 8821–8831. Pmlr, 2021. 2

    2021

  48. [56]

    Hierarchical text-conditional image gener- ation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gener- ation with clip latents. ArXiv, abs/2204.06125, 2022. 1, 2

    2022 arXiv

  49. [57]

    Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer

    Robin Rombach, A. Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10674–10685, 2021. 1, 2

    2022

  50. [58]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2

    2022

  51. [59]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. Advances in neural information...

    2022

  52. [60]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. arXiv preprint arXiv:2202.00512, 2022. 1, 2

    2022 arXiv

  53. [61]

    Adversarial diffusion distillation

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distillation. arXiv preprint arXiv:2311.17042, 2023. 1, 3, 6, 7, 5

    2023 arXiv

  54. [62]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural In- f...

    2022

  55. [63]

    High-dimensional continuous con- trol using generalized advantage estimation

    John Schulman, Philipp Moritz, Sergey Levine, Michael Jor- dan, and Pieter Abbeel. High-dimensional continuous con- trol using generalized advantage estimation. arXiv preprint arXiv:1506.02438, 2015. 2

    2015 arXiv

  56. [64]

    Proximal policy optimization algo- rithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Rad- ford, and Oleg Klimov. Proximal policy optimization algo- rithms. arXiv preprint arXiv:1707.06347, 2017. 1, 4, 7, 2

    2017 arXiv

  57. [65]

    Defining and characterizing reward gam- ing

    Joar Skalse, Nikolaus Howe, Dmitrii Krasheninnikov, and David Krueger. Defining and characterizing reward gam- ing. Advances in Neural Information Processing Systems , 35:9460–9471, 2022. 5

    2022

  58. [66]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International confer- ence on machine learning, pages 2256–2265. PMLR, 2015. 1, 2, 3

    2015

  59. [67]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 1

    2010 arXiv

  60. [68]

    Generative modeling by esti- mating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution. Advances in neural information processing systems, 32, 2019. 1, 2

    2019

  61. [69]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. arXiv preprint arXiv:2011.13456, 2020. 1, 2

    2011 arXiv

  62. [70]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. arXiv preprint arXiv:2303.01469, 2023. 1, 2, 3

    2023 arXiv

  63. [71]

    Learning to predict by the methods of temporal differences

    Richard S Sutton. Learning to predict by the methods of temporal differences. Machine learning, 3:9–44, 1988. 6

    1988

  64. [72]

    Reinforcement learning: An introduction

    Richard S Sutton. Reinforcement learning: An introduction. A Bradford Book, 2018. 3, 6

    2018

  65. [73]

    Policy gradient methods for reinforcement learning with function approximation

    Richard S Sutton, David McAllester, Satinder Singh, and Yishay Mansour. Policy gradient methods for reinforcement learning with function approximation. Advances in neural information processing systems, 12, 1999. 4

    1999

  66. [74]

    Evalalign: Supervised fine- tuning multimodal llms with human-aligned data for evalu- ating text-to-image models

    Zhiyu Tan, Xiaomeng Yang, Luozheng Qin, Mengping Yang, Cheng Zhang, and Hao Li. Evalalign: Supervised fine- tuning multimodal llms with human-aligned data for evalu- ating text-to-image models. CoRR, 2024. 1, 3

    2024

  67. [75]

    Understanding reinforcement learning-based fine-tuning of diffusion models: A tutorial and review

    Masatoshi Uehara, Yulai Zhao, Tommaso Biancalani, and Sergey Levine. Understanding reinforcement learning-based fine-tuning of diffusion models: A tutorial and review. arXiv preprint arXiv:2407.13734, 2024. 6

    2024 arXiv

  68. [76]

    Diffusion model align- ment using direct preference optimization

    Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik. Diffusion model align- ment using direct preference optimization. In Proceedings of the IEEE/CVF Conference on Computer Vision a...

    2024

  69. [77]

    Diffu- siondb: A large-scale prompt gallery dataset for text-to- image generative models

    Zijie J Wang, Evan Montoya, David Munechika, Haoyang Yang, Benjamin Hoover, and Duen Horng Chau. Diffu- siondb: A large-scale prompt gallery dataset for text-to- image generative models. arXiv preprint arXiv:2210.14896,

    arXiv

  70. [78]

    Q-learning

    Christopher JCH Watkins and Peter Dayan. Q-learning. Ma- chine learning, 8:279–292, 1992. 2, 3, 6

    1992

  71. [79]

    Better aligning text-to-image models with human preference

    Xiaoshi Wu, Keqiang Sun, Feng Zhu, Rui Zhao, and Hong- sheng Li. Better aligning text-to-image models with human preference. ArXiv, abs/2303.14420, 2023. 3

    2023 arXiv

  72. [80]

    Human preference score: Better aligning text- to-image models with human preference

    Xiaoshi Wu, Keqiang Sun, Feng Zhu, Rui Zhao, and Hong- sheng Li. Human preference score: Better aligning text- to-image models with human preference. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 2096–2105, 2023. 1, 3, 5, 7

    2023

  73. [81]

    Deep reward supervisions for tuning text-to-image diffusion models

    Xiaoshi Wu, Yiming Hao, Manyuan Zhang, Keqiang Sun, Zhaoyang Huang, Guanglu Song, Yu Liu, and Hongsheng Li. Deep reward supervisions for tuning text-to-image diffusion models. arXiv preprint arXiv:2405.00760, 2024. 3

    2024 arXiv

  74. [82]

    Group normalization

    Yuxin Wu and Kaiming He. Group normalization. In Pro- ceedings of the European conference on computer vision (ECCV), pages 3–19, 2018. 3

    2018

  75. [83]

    Florence-2: Advancing a unified representation for a variety 11 of vision tasks

    Bin Xiao, Haiping Wu, Weijian Xu, Xiyang Dai, Houdong Hu, Yumao Lu, Michael Zeng, Ce Liu, and Lu Yuan. Florence-2: Advancing a unified representation for a variety 11 of vision tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages ...

    2024

  76. [84]

    Imagere- ward: Learning and evaluating human preferences for text- to-image generation

    Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagere- ward: Learning and evaluating human preferences for text- to-image generation. Advances in Neural Information Pro- cessing Systems, 36, 2024. 1, 2, 3, 5, 6, 8

    2024

  77. [85]

    Using human feedback to fine-tune diffusion models without any reward model

    Kai Yang, Jian Tao, Jiafei Lyu, Chunjiang Ge, Jiaxin Chen, Weihan Shen, Xiaolong Zhu, and Xiu Li. Using human feedback to fine-tune diffusion models without any reward model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8941– 895...

    2024

  78. [86]

    Lion: La- tent point diffusion models for 3d shape generation

    Xiaohui Zeng, Arash Vahdat, Francis Williams, Zan Goj- cic, Or Litany, Sanja Fidler, and Karsten Kreis. Lion: La- tent point diffusion models for 3d shape generation. ArXiv, abs/2210.06978, 2022. 2

    2022 arXiv

  79. [87]

    Unifl: Improve stable diffusion via unified feedback learn- ing

    Jiacheng Zhang, Jie Wu, Yuxi Ren, Xin Xia, Huafeng Kuang, Pan Xie, Jiashi Li, Xuefeng Xiao, Min Zheng, Lean Fu, et al. Unifl: Improve stable diffusion via unified feedback learn- ing. arXiv preprint arXiv:2404.05595, 2024. 3, 5

    2024 arXiv

  80. [88]

    Bertscore: Evaluating text genera- tion with bert

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Wein- berger, and Yoav Artzi. Bertscore: Evaluating text genera- tion with bert. arXiv preprint arXiv:1904.09675, 2019. 8 12 Appendix A. Empirical Local Lipchitz of Two-step LCMs 0.075 0.100 0.125 0.150 20 50 100 200 500 with Ge...

    1904 arXiv

  81. [89]

    = ∥D(E(D(zt 2))) − D(zt 2)∥ ℓrec(zt 2(ϵ)) = ∥D(E(D(zt 2(ϵ)))) − D(zt 2(ϵ))∥ and for r we have: r(zt 2, c) = r(D(zt 2), c) r(zt 2(ϵ), c) = r(D(zt 2(ϵ)), c) Combined, we can then compute the two empirical local Lipschitz constants given the input data’s noise level t as: Llocal(...

  82. [90]

    − ℓrec(zt 2(ϵ))| ∥zt 1 − zt 1(ϵ)∥ Llocal(r, t) = E(zτ0 ,c) |r(zt 2, c) − r(zt 2(ϵ), c)| ∥zt 1 − zt 1(ϵ)∥ 1 Both empirical Lipchitz are estimated over N = 1000 sampled (zτ0 , c) pairs. To visualize the growing non-smoothness of the mapping when the input data has a larger noise...

  83. [91]

    Specifically, we use the CLIP-ViT-L/14 from the original CLIP paper and a similar BLIP model fine-tuned for VQA tasks (ViT-L for the vision backbone)

    or BLIP [35] to validate our choices. Specifically, we use the CLIP-ViT-L/14 from the original CLIP paper and a similar BLIP model fine-tuned for VQA tasks (ViT-L for the vision backbone). We train both in the latent space of SDXL as a fair comparison with our design in LaSRO....

  84. [92]

    Note that, SDXL-Turbo mainly focuses on image generation of 5122 pixels and SDXL-Lightning only supports≥ 2 step generation

    (2-step) and SDXL-Turbo [61] (1- and 2-step), both based on adversarial distillation. Note that, SDXL-Turbo mainly focuses on image generation of 5122 pixels and SDXL-Lightning only supports≥ 2 step generation. In addition, we examine applying LaSRO toward improving SDXL-Turbo...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.