DiFA: Inference-Time Forward-Process Alignment for Diffusion Models

Delu Zeng; Shigui Li

REVIEW 3 major objections 6 minor 12 references

DiFA: Inference-Time Forward-Process Alignment for Diffusion Models

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read DiFA claims that diffusion sampling quality can be improved for free by treating each denoiser prediction as a noisy observation of a common clean anchor, fusing recent predictions into a noise-level-weighted consensus, and adding back guid

desk verdict Training-free prediction-refinement wrapper with big reported FID gains at few steps; the Kalman framing overclaims, but the empirical package is new and deserves a serious referee. read the letter →

arxiv 2607.17972 v1 pith:6GAEJJJX submitted 2026-07-20 cs.LG

Shigui Li , Delu Zeng This is my paper · ORCID

classification cs.LG

keywords diffusionmodelsinference-timerefinementtemporalconsensusKalmanfilteringfew-stepsamplingpredictionalignmentFIDimprovementtraining-free

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 argues that the standard view of diffusion inference as numerical integration is incomplete: denoiser outputs are not exact estimates but noisy, biased readings of a locally stable clean anchor. DiFA therefore reframes inference-time refinement as sequential state estimation, building a temporal consensus from recent predictions weighted by signal-to-noise ratio, and then re-injecting residual detail via deviation guidance. If correct, this means existing pretrained diffusion and flow-matching models contain untapped self-correction capacity that can be unlocked with a training-free wrapper and negligible compute. The headline evidence is consistent improvement in FID, IS, and FD-DINOv2 across CIFAR-10, ImageNet, LSUN Bedroom, and SiT-XL, with the largest gains in the few-step regime.

What carries the argument

The load-bearing object is the forward-aligned temporal consensus: a weighted average of recently produced denoiser predictions, with weights derived from the forward process's inverse-SNR observation covariance (Proposition 4.2 and its recursive Kalman equivalent, Theorem 4.3). In practice the consensus is computed in a causal window of size three, using channel-wise affine alignment, structural similarity after pooling, and logSNR proximity; the deviation between the current prediction and this anchor is then orthogonally projected and split into low- and high-frequency components gated by logSNR before being added back with strength omega. This machinery lets DiFA refine the clean predict

What would settle it

Measure the empirical covariance of clean predictions from consecutive denoiser calls along a real few-step trajectory (e.g., the 3-step window DiFA uses). If the off-diagonal correlation of the prediction errors is large (say above roughly 0.5) or the mean squared distance between consecutive predictions is comparable to the variance DiFA's consensus claims to remove, then the BLUE/Kalman motivation collapses, and the reported FID gains would need a different explanation.

Watch

Extended reading notes

Core claim

The central claim is that the clean-signal predictions produced along a reverse diffusion trajectory can be treated as correlated observations of a trajectory-implied clean anchor, with forward-process noise-level structure providing the reliability ordering. Under an idealized independent-view model, the best linear unbiased estimator of the anchor is a precision-weighted average with weights proportional to SNR(t), equivalent to a static-state Kalman filter. DiFA instantiates this as a causal sliding-window consensus over recent predictions, aligned by channel statistics and weighted by structural similarity and logSNR proximity, then adds a modulated anchor-relative deviation to preserve

Load-bearing premise

The whole variance-reduction argument rests on treating recent denoiser predictions as independent, unbiased measurements of one fixed clean image with noise variance c/SNR(t); in reality those predictions come from the same network and solver, so they are correlated, biased, and may point at a moving target.

Editorial extensions

If this is right

No extra network evaluations are needed: DiFA only maintains a small buffer and computes O(Kd) per step, so it is a plug-and-play wrapper for existing solvers.
Largest gains come in the few-step regime: e.g., DPM-Solver++ on CIFAR-10 at 8 NFE improves from FID 8.40 to 4.15, and Heun at 5 NFE on ImageNet-64 improves from 230.05 to 110.20.
Improvements are consistent across FID, IS, and FD-DINOv2, suggesting the gain is not merely distributional smoothing.
The principle generalizes beyond pixel-space diffusion: latent diffusion on LSUN Bedroom and flow matching on SiT-XL/2 also improve.
At higher step counts gains shrink but remain positive: CIFAR-10 at 20 NFE goes from 2.33 to 1.96, and ImageNet-64 at 25 NFE reaches 1.63–1.64 FID.
DiFA is solver-compatible: it works with DDIM, DPM-Solver++, UniPC, and Heun, and can be adapted to flow-matching trajectories.

Reading between the lines

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

If the consensus works because MSE-trained denoisers share a common systematic error across steps, the same filtering could be applied to other iterative generative models whose outputs are noisy estimates of a shared latent, such as autoregressive or energy-based samplers.
The paper's own idealized model suggests a diagnostic: the size of the FID gain should correlate with the empirical correlation of consecutive prediction errors. Measuring that correlation on a few-step run would reveal whether the gain is true variance reduction or instead implicit smoothing.
A potential failure mode follows from the model: when the trajectory-implied anchor drifts within the window (high curvature, very low NFE), the consensus averages different targets. In that regime a smaller window or an explicit anchor-motion term should be needed; this is testable and not explored in the paper.
Because DiFA only changes the clean prediction, it could in principle be combined with distillation or one-step methods, although the paper does not test that combination; the gain might vanish once trajectory predictions are already transported.

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.

Desk Editor's Note

read the letter

DiFA is a training-free wrapper that refines the clean prediction at each diffusion/flow sampling step using a causal temporal consensus over previous predictions plus a residual 'deviation guidance' term. The headline claim is real and novel as a package: plug it into DDIM, DPM-Solver++, UniPC, Heun, or a flow-matching sampler and FID/IS/FD-DINOv2 improve consistently, often dramatically at 5–8 NFE, at no extra network evaluations. That is a useful result if it holds.

What the paper does well: the experiments are broad — CIFAR-10, ImageNet-64, LSUN Bedroom, SiT-XL/2 with and without CFG — and the ablations isolate the roles of window size, SNR gating, magnitude alignment, and scale. They also report FD-DINOv2, which is easy to omit. The cross-paradigm transfer to flow matching is a nice sanity check. The method itself is simple to implement and the appendix gives the exact equations for the default instantiation.

The soft spots are real. The BLUE/Kalman analysis (Prop. 4.2, Thm. 4.3) proves variance reduction only under an idealized observation model — independent, unbiased observations with covariance c/SNR(t)I. The implemented estimator is a softmax over structural-similarity and logSNR compatibility of channel-wise affine-aligned historical predictions, and the final refined prediction is the current prediction plus a frequency-gated residual. The authors admit in Sec. 4.1, A.1, and the Conclusion that actual predictions are biased and correlated, so the theory is motivation, not certification. That is fine if presented as such, but the paper leans on the Kalman framing more than the algorithm supports.

The empirical case has two gaps. First, no error bars or multiple seeds are reported anywhere. At NFE 20 on CIFAR-10 the margin is 1.96 vs 2.33; on ImageNet-64 at 25 NFE it's 1.63–1.64 vs 1.73–1.83. Those are precisely the regimes where FID seed variance can matter. Second, several hyperparameters are tuned in-sample — Table 14 reports the best scale for each NFE, and the main text selects the best configuration. That's not unusual for a method paper, but it means the headline numbers are optimistic until confirmed independently. Also, some implementation details are vague: the code link has no hash, and mu(logSNR), epsilon stabilizers, and phi variants are described with 'set according to the implementation configuration' rather than pinned down.

Overall: the central claim is plausible and the experiments are consistent. The theory should either be aligned with the implemented algorithm or explicitly labeled as intuition. The paper deserves a serious referee. A revision with error bars, fixed hyperparameters selected on a validation set, and a code release with a hash would substantially raise my confidence. Would bring to reading group; I'd cite it if I work on inference-time refinement.

Referee Report

3 major / 6 minor

Summary. The paper proposes DiFA, a training-free, solver-agnostic wrapper that refines each clean-signal prediction during diffusion or flow-matching inference. It maintains a causal buffer of K historical predictions, aligns them channel-wise to the current prediction, forms a consensus anchor using softmax weights over structural similarity and logSNR proximity, and adds a modulated residual (orthogonal projection, low/high-frequency gating) to the current prediction. The paper claims consistent FID/IS/FD-DINOv2 improvements across CIFAR-10, ImageNet-64, LSUN Bedroom, and SiT-XL ImageNet-256, with no additional network evaluations. The theoretical sections (Prop. 4.2, Thm. 4.3) derive BLUE/Kalman optimality under an idealized observation model (Eq. 13), which the paper itself concedes is not an exact description of practical denoiser predictions.

Significance. If the empirical results are reproducible, DiFA would be a valuable, low-cost plug-in for existing samplers, with large margins in the few-step regime (e.g., CIFAR-10 8-NFE DPM-Solver++ FID 8.40 to 4.15; SiT-XL Euler 5-NFE FID 52.64 to 27.61). The method introduces no extra NFEs and has negligible per-step overhead. The paper is commendably explicit about the idealization in its theoretical model and includes extensive ablations. However, the theory as stated does not cover the implemented estimator, and the empirical claims would be substantially strengthened by uncertainty quantification and a fixed hyperparameter-selection protocol. The core idea is promising but currently overstates its principled grounding.

major comments (3)

[§4.2–§4.4, App. A.3 (Eqs. 13–15, 46–49, 22)] The BLUE/Kalman derivation (Prop. 4.2, Thm. 4.3) applies only to the idealized model Eq. (13): independent observations of a static anchor with covariance c/SNR(t) I. The implemented estimator is structurally different: the consensus weights in Eq. (49) are softmax over τ·m̄_ij − μ|ℓ_i−ℓ_j|, not the precision weights of Eq. (15); the current prediction is excluded from the consensus (Sec. 4.4); and the final refined prediction in Eq. (22) is the current prediction plus a modulated residual, not the BLUE estimate. The paper concedes this in App. A.1 ('practical denoiser predictions are organized according to the resulting anchor-consistency and reliability-ordering principles, rather than being assumed to satisfy the same covariance model exactly'). As a result, Prop. 4.2 and Thm. 4.3 do not certify variance reduction for the actual DiFA output. The theory should be either aligned with th
[§4.1, Eq. (10); Conclusion] The observation model Eq. (10) treats historical predictions as a locally stable anchor plus bias b_j|i and residual ξ_j|i, but provides no analysis of temporal correlation or anchor drift. In practice, predictions along a reverse trajectory are deterministically coupled through the solver state and the same network, so they are strongly correlated and biased. The paper's own Conclusion states that the framework relies on 'an idealized static-anchor assumption' and defers 'theoretical extensions for temporally correlated errors' to future work. This is an explicit limitation of the central claim that DiFA is 'forward-process aligned.' Please provide either a quantitative drift/correlation analysis for the K=3 window, especially in high-curvature few-step regimes, or soften the claim that the forward geometry is the source of the empirical gains.
[§5.2, Table 14; App. B.3] Hyperparameters appear to be selected in-sample. Table 14 reports, for each solver and NFE budget, the best FID among four refinement scales (1.25, 1.5, 1.7, 1.75), and the main-text comparisons in Tables 1–2 use the default s=1.7 but the appendix sweeps many parameters (window size, SNR threshold, scale, compatibility). The reported margins at larger NFE are small (e.g., ImageNet-64 25-NFE 1.64 vs 1.83; CIFAR-10 20-NFE 1.96 vs 2.33), where FID variability across seeds can be material. I request error bars or multiple-seed results, and either a single fixed hyperparameter set used across all NFE/solver combinations or an explicit validation-selection protocol that avoids picking the best per cell. Without this, the 'consistently improves' claim is not fully robust.

minor comments (6)

[General] The code link is given as 'DiFA' with no URL; include a working anonymous link for review.
[Introduction] Typo: 'V AEs' should be 'VAEs'.
[Figure 6] The log-scale y-axis makes FID curves difficult to read; consider annotating key values directly on the curves or using a linear scale for the low-FID region.
[Tables 1–2] The 'Ours' rows list only the DiFA result; including the baseline FID in the same row would make the comparison self-contained.
[App. B.2, Table 7] The 'w/o History' row sets W=1 and reports identical results to the naive baseline; this is expected but should be stated explicitly in the text for clarity.
[App. A.3, Eq. (45)] Channel-wise affine alignment is applied before computing similarity; the effect of this alignment on the subsequent softmax weights is not discussed. A brief comment on why alignment does not erase the signal being compared would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DiFA's BLUE/Kalman derivation is explicitly an idealized motivation, and the headline FID/IS results are external benchmark evaluations rather than consequences of a self-referential definition.

full rationale

The paper separates the ideal forward-aligned observation model (Eq. 13) from the practical consensus (Eq. 19, Appendix A.3 Eq. 49). The BLUE/Kalman variance-reduction results (Prop. 4.2, Thm. 4.3) are presented as motivation, not as a proof about the implemented estimator; Eq. (15) is not defined in terms of Eq. (49), and the practical weights are not claimed to be the precision weights. The main claim—consistent FID/IS/FD-DINOv2 improvement over base samplers—is evaluated on external benchmarks with fixed pretrained checkpoints and reported alongside ablations; this is empirical evidence, not a derivation that reduces to its inputs. Self-citations (e.g., Li & Zeng 2026 for cumulative drift; EVODiff as prior work) are contextual and are not load-bearing for the DiFA update. The acknowledged idealization of independent unbiased observations and the benchmark-tuned hyperparameters are limitations and overfitting risks, not circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced. The 'forward-aligned consensus anchor' is a computed algorithmic quantity, not a postulate. The framing rests on a static-anchor independence assumption that is explicitly said to be an analytical proxy. Free parameters include the window size, softmax sharpness, deviation scale, pooling kernels, the unspecified logSNR-compatibility function, and dataset-specific SNR gating/magnitude factors.

free parameters (7)

K (causal temporal window size) = 3
History buffer length; ablation (Table 8) shows U-shaped FID, default 3 chosen as best.
tau (structural compatibility sharpness) = 4.0
Softmax temperature over similarity scores; default chosen, no sensitivity table for tau in main setting.
s (deviation scale, omega=s-1) = 1.7 (omega=0.7)
Deviation guidance strength; sweep (Tables 11, 14) shows best around 1.7-1.75; fitted on benchmark FID.
k_sim (similarity pooling kernel) = 3
Average-pooling kernel for similarity computation; hand-chosen.
k_res (residual smoothing kernel) = 5
Residual smoothing kernel; hand-chosen.
mu(logSNR) logSNR compatibility function = unspecified
Function controlling penalty |logSNR_i - logSNR_j|; exact form unspecified in manuscript ('set according to the implementation configuration').
SNR_gating threshold and magnitude alignment phi (EDM/LDM variants) = SNR_lo=-2.5, phi=0.5 (EDM 10-NFE ablation)
SNR gating threshold and magnitude alignment factor used in EDM/LDM variants (Tables 5, 9, 10); fitted to FID.

assumptions (4)

domain assumption Historical denoiser predictions are modeled as conditionally independent observations of a locally static clean anchor with covariance (c/SNR(t)) I (Eq. 13).
This ideal model is used to derive BLUE/Kalman consensus and Prop 4.2; actual predictions are coupled, biased and correlated, as the paper concedes in §4.1 and A.1.
domain assumption The trajectory-implied clean anchor x_0,*^(t_i) is locally stable within the causal window W_K(t_i) (Eq. 10).
If the anchor drifts during few-step inference, averaging over the window averages distinct targets and introduces bias; no analysis of anchor drift is provided.
ad hoc to paper Softmax weights over structural cosine similarity minus logSNR distance (Eqs. 46-49) form a valid fusion rule for historical predictions.
The heuristic weighting scheme is not derived from the forward process or from the BLUE/Kalman analysis; its success is empirical only.
ad hoc to paper Orthogonal residual projection and sigmoid high-frequency gating (Eqs. 52-55) recover useful details without destabilizing sampling.
Deviation guidance is designed by hand; its behavior on correlated prediction errors has no theoretical guarantee.

how reviews work

0 comments

Cite this review

Pith. "Pith review of DiFA: Inference-Time Forward-Process Alignment for Diffusion Models." pith.science (2026). https://pith.science/paper/6GAEJJJX

@misc{pith2026260717972,
  author       = {Pith},
  title        = {Pith review of: DiFA: Inference-Time Forward-Process Alignment for Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6GAEJJJX}},
  note         = {Machine review of arXiv:2607.17972}
}

read the original abstract

The prevailing inference framework for diffusion models formulates generation fundamentally as a problem of numerical integration. This perspective casts the model as an exact estimator, neglecting the inherent statistical uncertainty of the denoising process. In this work, we propose Forward-Process Aligned Diffusion prediction (\textbf{DiFA}), a training-free framework that reframes inference-time data prediction refinement as a sequential state estimation problem. Rather than reusing past outputs solely for numerical integration, DiFA treats iterative data predictions along the reverse trajectory as correlated observations to build a forward-aligned temporal consensus. Inspired by Kalman filtering, this consensus aggregates historical predictions according to structural consistency and noise-level compatibility. To counteract the over-smoothing tendency of temporal consensus, we introduce a deviation guidance mechanism to adaptively preserve residual details. Empirically, DiFA yields significant improvements on CIFAR-10 and ImageNet across the evaluated metrics, including FID, IS, and FD-DINOv2, demonstrating that aligning inference with the forward statistical structure substantially improves generative fidelity.

Figures

Figures reproduced from arXiv: 2607.17972 by the authors.

**Figure 2.** Illustrating the mean-field discrepancy in model estimation through spatial shift and magnitude scaling. (PF-ODE) that shares the same marginal as the SDE: d x d t = f(t)x − 1 2 g(t) 2∇x log pt(x). (4) 3.2. Inference Framework and Linearization Bias As the score function ∇x log pt(x) is parameterized as −ϵθ(xt, t)/σt (Song et al., 2021c; Karras et al., 2022) and the prediction relationship of xθ (xt, t) = xt−σtϵθ(x… view at source ↗

**Figure 3.** Visual quality comparison on CIFAR-10 using pretrained EDM weights. Compared to the baseline solver, incorporating our DiFA framework significantly enhances visual fidelity and edge sharpness under identical few-step inference budgets. For i = 2, . . . , n, assimilate yi using the static-state Kalman update with observation covariance c SNR(ti) −1I. Then xˆ rec 0,n = Pn i=1 P SNR(ti)yi n i=1 SNR(ti) = xˆ ⋆ 0,ideal… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Visual quality comparison of generated samples on ImageNet 256 × 256 using the pre-trained SiT-XL/2 with a classifier-free guidance (CFG) scale of 1.5. SiT is a DiT-based architecture trained via Flow Matching. We compare the baseline Euler and Heun samplers against th…

**Figure 5.** Figure 5: Perceptual Quality (FD-DINOv2). DiFA also yields consistent improvements in FD-DINOv2 scores, indicating that our method enhances not just statistical fidelity (FID) but also perceptually aligned image quality. effective under both logSNR and EDM noise schedules, sugge…

**Figure 6.** Figure 6: Quantitative Results. Top Row: On CIFAR-10, DiFA (solid red lines) significantly lowers FID compared to baselines (dashed gray lines) across all NFE regimes for both DPM-Solver++ and UniPC. Note the rapid convergence at NFE 12 (FID ≈ 2.2). Bottom Row: On ImageNet 64x64…

**Figure 7.** Figure 7: Qualitative comparison on ImageNet (256 × 256) using the pretrained SiT-XL/2 model with the Euler solver, 5 function evaluations (NFE = 5), and a classifier-free guidance (CFG) scale of 4.0. Compared with the baseline, DiFA preserves better structural coherence and loc…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 5 linked inside Pith

[4]

Fu, S., Wang, Y ., Chen, Y ., Shen, L., and Tao, D

URL https://openreview.net/forum? id=OlzB6LnXcS. Fu, S., Wang, Y ., Chen, Y ., Shen, L., and Tao, D. Self- verification provably prevents model collapse in recur- sive synthetic training. In Belgrave, D., Zhang, C., Lin, H., Pascanu, R., Koniusz, P., Ghassemi, M., and Chen, N. (eds.),Advances in Neural Information Pro- cessing Systems, volume 38, pp. 3610...

arXiv 2025
[5]

Karras, T., Aittala, M., Hellsten, J., Laine, S., Lehtinen, J., and Aila, T

URL https://openreview.net/forum? id=ThMQfsBnje. Karras, T., Aittala, M., Hellsten, J., Laine, S., Lehtinen, J., and Aila, T. Training generative adversarial networks with limited data. InAdvances in Neural Information Processing Systems, volume 33, pp. 12104–12114, 2020. Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of dif...

arXiv 2020
[6]

Li, W., Song, Z., Zhou, H., Yu, J., Zhang, Y ., and Yang, W

URL https://openreview.net/forum? id=UXEo3uNNIX. Li, W., Song, Z., Zhou, H., Yu, J., Zhang, Y ., and Yang, W. LoRA-mixer: Coordinate modular loRA experts through serial attention routing. InThe Fourteenth In- ternational Conference on Learning Representations,
[7]

Lin, S., Liu, B., Li, J., and Yang, X

URL https://openreview.net/forum? id=GMP1S4R6Ke. Lin, S., Liu, B., Li, J., and Yang, X. Common diffusion noise schedules and sample steps are flawed. InProceedings of the IEEE/CVF winter conference on applications of computer vision, pp. 5404–5411, 2024. Lipman, Y ., Chen, R. T. Q., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modelin...

2024
[9]

Podell, D., English, Z., Lacey, K., Blattmann, A., Dock- horn, T., M¨uller, J., Penna, J., and Rombach, R

URL https://openreview.net/forum? id=TubfvqZNEr. Podell, D., English, Z., Lacey, K., Blattmann, A., Dock- horn, T., M¨uller, J., Penna, J., and Rombach, R. SDXL: Improving latent diffusion models for high-resolution image synthesis. InThe Twelfth International Confer- ence on Learning Representations, 2024. URL https: //openreview.net/forum?id=di52zR8xgf....

2024
[10]

Sabour, A., Fidler, S., and Kreis, K

URL https://openreview.net/forum? id=nBGBzV4It3. Sabour, A., Fidler, S., and Kreis, K. Align your flow: Scal- ing continuous-time flow map distillation. InThe Thirty- ninth Annual Conference on Neural Information Pro- cessing Systems, 2025. URL https://openreview. net/forum?id=pzHuesCvcO. Salimans, T. and Ho, J. Progressive distillation for fast sampling ...

arXiv 2025
[11]

Zhang, Q

URL https://openreview.net/forum? id=MtDd7rWok1. Zhang, Q. and Chen, Y . Fast sampling of diffusion mod- els with exponential integrator. InThe Eleventh In- ternational Conference on Learning Representations,
[2022]

Liu, X., Gong, C., and qiang liu

URL https://openreview.net/forum? id=PlKWVd2yBkY. Liu, X., Gong, C., and qiang liu. Flow straight and fast: Learning to generate and transfer data with rec- tified flow. InThe Eleventh International Conference on Learning Representations, 2023a. URL https: //openreview.net/forum?id=XVjTT1nw5z. Liu, X., Park, D. H., Azadi, S., Zhang, G., Chopikyan, A., Hu,...

arXiv 2024

Show all 12 references

[2023]

Zhao, W., Bai, L., Rao, Y ., Zhou, J., and Lu, J

URL https://openreview.net/forum? id=Loek7hfb46P. Zhao, W., Bai, L., Rao, Y ., Zhou, J., and Lu, J. UniPC: A unified predictor-corrector framework for fast sampling of diffusion models. InThirty-seventh Conference on Neural Information Processing Systems, 2023. URL https: //op...

2023 arXiv
[2024]

Bao, F., Li, C., Zhu, J., and Zhang, B

URL https://openreview.net/forum? id=pzpWBbnwiJ. Bao, F., Li, C., Zhu, J., and Zhang, B. Analytic-DPM: an analytic estimate of the optimal reverse variance in diffusion probabilistic models. InInternational Confer- ence on Learning Representations, 2022. URL https: //openrevie...

2022
[2025]

Bansal, A., Chu, H.-M., Schwarzschild, A., Sengupta, R., Goldblum, M., Geiping, J., and Goldstein, T

URL https://openreview.net/forum? id=MKvQH1ekeY. Bansal, A., Chu, H.-M., Schwarzschild, A., Sengupta, R., Goldblum, M., Geiping, J., and Goldstein, T. Uni- versal guidance for diffusion models. InThe Twelfth International Conference on Learning Representations,
[2026]

doi: https://doi.org/10.1016/j

ISSN 0031-3203. doi: https://doi.org/10.1016/j. patcog.2025.112442. Chen, Z., Li, H., Wang, F., Zhang, O., Xu, H., Jiang, X., Song, Z., and Wang, H. Rethinking the diffusion models for missing data imputation: A gradient flow perspective. Advances in Neural Information Process...

2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [4]

Fu, S., Wang, Y ., Chen, Y ., Shen, L., and Tao, D

URL https://openreview.net/forum? id=OlzB6LnXcS. Fu, S., Wang, Y ., Chen, Y ., Shen, L., and Tao, D. Self- verification provably prevents model collapse in recur- sive synthetic training. In Belgrave, D., Zhang, C., Lin, H., Pascanu, R., Koniusz, P., Ghassemi, M., and Chen, N. (eds.),Advances in Neural Information Pro- cessing Systems, volume 38, pp. 3610...

arXiv 2025

[2] [5]

Karras, T., Aittala, M., Hellsten, J., Laine, S., Lehtinen, J., and Aila, T

URL https://openreview.net/forum? id=ThMQfsBnje. Karras, T., Aittala, M., Hellsten, J., Laine, S., Lehtinen, J., and Aila, T. Training generative adversarial networks with limited data. InAdvances in Neural Information Processing Systems, volume 33, pp. 12104–12114, 2020. Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of dif...

arXiv 2020

[3] [6]

Li, W., Song, Z., Zhou, H., Yu, J., Zhang, Y ., and Yang, W

URL https://openreview.net/forum? id=UXEo3uNNIX. Li, W., Song, Z., Zhou, H., Yu, J., Zhang, Y ., and Yang, W. LoRA-mixer: Coordinate modular loRA experts through serial attention routing. InThe Fourteenth In- ternational Conference on Learning Representations,

[4] [7]

Lin, S., Liu, B., Li, J., and Yang, X

URL https://openreview.net/forum? id=GMP1S4R6Ke. Lin, S., Liu, B., Li, J., and Yang, X. Common diffusion noise schedules and sample steps are flawed. InProceedings of the IEEE/CVF winter conference on applications of computer vision, pp. 5404–5411, 2024. Lipman, Y ., Chen, R. T. Q., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modelin...

2024

[5] [9]

Podell, D., English, Z., Lacey, K., Blattmann, A., Dock- horn, T., M¨uller, J., Penna, J., and Rombach, R

URL https://openreview.net/forum? id=TubfvqZNEr. Podell, D., English, Z., Lacey, K., Blattmann, A., Dock- horn, T., M¨uller, J., Penna, J., and Rombach, R. SDXL: Improving latent diffusion models for high-resolution image synthesis. InThe Twelfth International Confer- ence on Learning Representations, 2024. URL https: //openreview.net/forum?id=di52zR8xgf....

2024

[6] [10]

Sabour, A., Fidler, S., and Kreis, K

URL https://openreview.net/forum? id=nBGBzV4It3. Sabour, A., Fidler, S., and Kreis, K. Align your flow: Scal- ing continuous-time flow map distillation. InThe Thirty- ninth Annual Conference on Neural Information Pro- cessing Systems, 2025. URL https://openreview. net/forum?id=pzHuesCvcO. Salimans, T. and Ho, J. Progressive distillation for fast sampling ...

arXiv 2025

[7] [11]

Zhang, Q

URL https://openreview.net/forum? id=MtDd7rWok1. Zhang, Q. and Chen, Y . Fast sampling of diffusion mod- els with exponential integrator. InThe Eleventh In- ternational Conference on Learning Representations,

[8] [2022]

Liu, X., Gong, C., and qiang liu

URL https://openreview.net/forum? id=PlKWVd2yBkY. Liu, X., Gong, C., and qiang liu. Flow straight and fast: Learning to generate and transfer data with rec- tified flow. InThe Eleventh International Conference on Learning Representations, 2023a. URL https: //openreview.net/forum?id=XVjTT1nw5z. Liu, X., Park, D. H., Azadi, S., Zhang, G., Chopikyan, A., Hu,...

arXiv 2024