# diffusers

A Python library for using pretrained diffusion models to generate videos, images, and audio.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Bigger Isn't Always Memorizing: Early Stopping Overparameterized Diffusion Models

DevFeed: [Bigger Isn't Always Memorizing: Early Stopping Overparameterized Diffusion Models](<https://devfeed.tech/articles/bigger-isn-t-always-memorizing-early-stopping-overparameterized-diffusion-models-143219.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2505.16959>)

Author: Alessandro Favero, Antonio Sclocchi, Matthieu Wyart

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Occam's razor machine learning](<https://devfeed.tech/topics/occam-s-razor-machine-learning.md>), [scaling laws](<https://devfeed.tech/topics/scaling-laws.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [competition](<https://devfeed.tech/tags/competition.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [diagram](<https://devfeed.tech/tags/diagram.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [grammar](<https://devfeed.tech/tags/grammar.md>), [image](<https://devfeed.tech/tags/image.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [language](<https://devfeed.tech/tags/language.md>), [noise](<https://devfeed.tech/tags/noise.md>), [optimize](<https://devfeed.tech/tags/optimize.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [probabilistic](<https://devfeed.tech/tags/probabilistic.md>), [random](<https://devfeed.tech/tags/random.md>), [regression](<https://devfeed.tech/tags/regression.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

In highly overparameterized diffusion models, generalization on natural data emerges during training before memorization begins; fitting training loss completely can lead to reproducing training examples. Across image and language models, the authors report that memorization time scales with dataset size, consistent with a kernel-regression bound on fitting the empirical score at low noise. They interpret generalization and memorization as competing time scales. A random-rule probabilistic context-free grammar model shows a similar pattern: deeper rules are acquired over time, and the authors characterize the generalization cost of stopping early. Their results support dataset-size-scaled early stopping as a way to improve generalization while avoiding memorization, with possible relevance to hyperparameter transfer and privacy-sensitive applications. The evidence provided is an abstract, so it does not specify a concrete stopping rule or experimental details.

### Source excerpt

arXiv:2505.16959v3 Announce Type: replace Abstract: Diffusion probabilistic models have become a cornerstone of modern generative AI, yet the mechanisms underlying their generalization remain poorly understood. In fact, if these models were perfectly minimizing their training loss, they would just generate data belonging to their training set, i.e., memorize, as empirically found in the overparameterized regime. We revisit this view by showing that, in highly overparameterized diffusion models, generalization in natural data domains is progressively achieved during training before the onset of memorization. Our results, ranging from image to language diffusion models, systematically support the empirical law that memorization time is proportional to the dataset size, consistent with a kernel-regression bound on the time required to fit the empirical score at low noise. Generalization vs. memorization is then best understood as a competition between time scales. We show that this phenomenology is recovered in diffusion models learning a simple probabilistic context-free grammar with random rules, where generalization corresponds to the hierarchical acquisition of deeper grammar rules as training time grows, and the generalization cost of early stopping can be characterized. We summarize these results in a phase diagram. Overall, our results support that a principled early-stopping criterion - scaling with dataset size - can effectively optimize generalization while avoiding memorization, with direct implications for hyperparameter transfer and privacy-sensitive applications.

## High-Dimensional Black-Box Optimization via Guidance Sampling of Diffusion Models

DevFeed: [High-Dimensional Black-Box Optimization via Guidance Sampling of Diffusion Models](<https://devfeed.tech/articles/high-dimensional-black-box-optimization-via-guidance-sampling-of-diffusion-models-140097.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22965>)

Author: Xiaoyan Hu, Rohit Sonker, Jalal Butt, Jiayu Chen, Egemen Kolemen, Jeff Schneider

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Operations research and optimization](<https://devfeed.tech/topics/operations-research-and-optimization.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>)

Tags: [bayesian](<https://devfeed.tech/tags/bayesian.md>), [black-box](<https://devfeed.tech/tags/black-box.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [discovery](<https://devfeed.tech/tags/discovery.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [exploitation](<https://devfeed.tech/tags/exploitation.md>), [exploration](<https://devfeed.tech/tags/exploration.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [generation](<https://devfeed.tech/tags/generation.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [guidance](<https://devfeed.tech/tags/guidance.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [sampling](<https://devfeed.tech/tags/sampling.md>)

### AI overview

BReaD uses diffusion-model guidance to optimize black-box rewards when evaluations are limited, without requiring a labeled pretraining dataset or access to reward gradients. It fits a Gaussian process surrogate to queried labels from generated samples, then uses Upper Confidence Bound signals to balance exploration and exploitation. The abstract reports a theoretical analysis of the total-variation distance between generated and target distributions, and numerical experiments on scientific discovery and image generation that show effectiveness in low-data scenarios.

### Source excerpt

arXiv:2611.22965v1 Announce Type: cross Abstract: Many real-world optimization problems involve optimizing black-box objectives for which only a limited number of evaluations are available. Bayesian optimization (BO) is a powerful framework for such problems, but its effectiveness often deteriorates in high-dimensional spaces. Recent work leverages diffusion models for black-box optimization, often through pre-training a reward-conditioned score network to steer generation towards high-reward samples. Alternatively, gradient-based guidance methods assume access to a differentiable reward function from which guidance gradients are queried. However, these approaches either require large amounts of labeled data for offline training or access to exact reward gradients, limiting their applicability in black-box settings. We propose BReaD (Black-box Reward-guided Sampling of Diffusion Model), a guidance sampling framework for optimizing black-box rewards without assuming access to a labeled dataset or reward gradients. The proposed method uses a Gaussian process (GP) as a differentiable surrogate for the unknown objective and actively queries labels for the generated samples. We provide a theoretical analysis of the total-variation distance between the generated distribution and the target distribution. Using Upper Confidence Bound (UCB) as guidance signals, our method balances exploration and exploitation during generation and enables the discovery of high-reward regions in the data space. Numerical experiments on scientific discovery and image generation tasks demonstrate that our framework can effectively optimize black-box rewards in low-data scenarios.

## Spectral Steering: Local Inverse Geometry for Frozen Diffusion Generators

DevFeed: [Spectral Steering: Local Inverse Geometry for Frozen Diffusion Generators](<https://devfeed.tech/articles/spectral-steering-local-inverse-geometry-for-frozen-diffusion-generators-137468.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19572>)

Author: Angelo Moroncelli, Alessandro Leanza, Loris Roveda

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [control](<https://devfeed.tech/tags/control.md>), [cosine](<https://devfeed.tech/tags/cosine.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generator](<https://devfeed.tech/tags/generator.md>), [generators](<https://devfeed.tech/tags/generators.md>), [guidance](<https://devfeed.tech/tags/guidance.md>), [images](<https://devfeed.tech/tags/images.md>), [local](<https://devfeed.tech/tags/local.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [motion](<https://devfeed.tech/tags/motion.md>), [planning](<https://devfeed.tech/tags/planning.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [scalar](<https://devfeed.tech/tags/scalar.md>), [steering](<https://devfeed.tech/tags/steering.md>)

### AI overview

Spectral steering aims to make structured edits to frozen diffusion generators while preserving other output properties. It predicts directional gains from the interaction between the request and the objective-specific Jacobian of the remaining sampler, then applies damped inverse correction with fixed-time relinearization for finite edits. In 8,923 instances from a retrospectively identified local regime, predicted and realized gains correlated at Spearman 0.79, compared with -0.11 for condition number, suggesting conditioning alone did not predict gains. The study covers images, audio, robotics, and control and planning. In 50 paired inpainting cases, four corrections improved unobserved-pixel PSNR by 3.25 dB over Adam at equal total derivative traversals. Scalar-null and matched-movement controls test the proposed mechanism.

### Source excerpt

arXiv:2611.19572v1 Announce Type: cross Abstract: A frozen diffusion generator may need to change several output properties while preserving others. We study when guidance faithfully realizes such a structured request and whether local inverse correction improves its direction. Our account links this benefit to the interaction between the request and the objective-specific remaining-sampler Jacobian: conditioning alone is insufficient. Spectral steering combines a request-specific cosine-gain predictor with damped inverse correction and fixed-time relinearization for finite edits. Scalar nulls and matched-movement controls test the predicted mechanism. Across 8,923 instances in a retrospectively identified local regime, predicted and realized gains correlate at Spearman 0.79, versus -0.11 for condition number. Experiments span images, audio, robotics, and control and planning, including complex character motion generation. On 50 paired inpainting cases, four corrections improve unobserved-pixel PSNR over Adam by 3.25 dB at equal total derivative traversals. Together, the analysis and controls connect request-dependent geometry to directional gains and establish where these gains translate into downstream performance.

## Do Heavy Tails Help Diffusion? On the Subtle Trade-off Between Initialization and Training

DevFeed: [Do Heavy Tails Help Diffusion? On the Subtle Trade-off Between Initialization and Training](<https://devfeed.tech/articles/do-heavy-tails-help-diffusion-on-the-subtle-trade-off-between-initialization-and-training-132786.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.13175>)

Author: Hamza Cherkaoui, H\\'el\\\`ene Halconruy, Antonio Ocello

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [flow](<https://devfeed.tech/tags/flow.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [models](<https://devfeed.tech/tags/models.md>), [noise](<https://devfeed.tech/tags/noise.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [tails](<https://devfeed.tech/tags/tails.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Heavy-tailed noise may speed initialization in diffusion models, but accumulated training errors can offset that advantage. The paper compares sampling-error bounds for two representative models using heavy-tailed and light-tailed Gaussian noise, then reports synthetic and real-world experiments supporting the theoretical findings. Its results question whether heavy-tailed noise reliably improves rare-region exploration.

### Source excerpt

arXiv:2605.13175v3 Announce Type: replace Abstract: Recent works have proposed incorporating heavy-tailed (HT) noise into diffusion- and flow-based generative models, with the goals of better recovering the tails of target distributions and improving generative diversity. This motivation is intuitive: if the data are heavy-tailed, HT noise may appear better matched than light-tailed (LT) Gaussian noise. However, replacing Gaussian noise by HT noise also changes the underlying estimation problem. In this paper, we revisit this paradigm through a combined theoretical and empirical study, establishing sampling-error bounds for two representative diffusion models driven by HT and LT noise. We show that HT diffusion models can achieve exponentially fast initialization, but accumulated training errors may offset this advantage. Experiments on synthetic and real-world datasets support our theoretical findings. Our results call into question a growing design trend in generative modeling and challenge the use of HT noise to improve rare-region exploration.

## DUO: Dual Readouts of Diffusion Models for OOD Detection

DevFeed: [DUO: Dual Readouts of Diffusion Models for OOD Detection](<https://devfeed.tech/articles/duo-dual-readouts-of-diffusion-models-for-ood-detection-128687.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13304>)

Author: Alireza Javid, Shirin Shoushtari, Nuria Gonz\\'alez Prelcic

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [outlier-detection](<https://devfeed.tech/topics/outlier-detection.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [ood](<https://devfeed.tech/tags/ood.md>)

### AI overview

DUO detects out-of-distribution inputs by combining diffusion-model reconstruction error with the leading sensitivity spectrum of intermediate denoising features. Reconstruction error alone can be unreliable because some OOD inputs may be reconstructed as accurately as, or better than, in-distribution inputs; feature sensitivity provides a complementary signal when that happens. The abstract describes DUO as fully unsupervised and training-free, reporting about a 6 percentage-point average AUROC improvement over the strongest baselines, substantially lower inference cost, and state-of-the-art results on near-OOD and high-resolution benchmarks.

### Source excerpt

arXiv:2611.13304v1 Announce Type: cross Abstract: Diffusion models are powerful generative methods for learning and sampling from complex data distributions. Safely deploying these models requires detecting inputs that fall outside their training distribution. Because diffusion models recover clean samples from noisy inputs, reconstruction error is often used for out-of-distribution (OOD) detection. This approach assumes that in-distribution (ID) inputs are reconstructed more accurately than OOD inputs. However, OOD inputs may be reconstructed equally well or even better, making this score unreliable. We therefore propose the leading sensitivity spectrum of intermediate denoising features as a complementary OOD signal and demonstrate that it remains informative when reconstruction error is misleading. We introduce DUal-readout of diffusion models for OOD detection (DUO), a fully unsupervised, training-free OOD detector that combines reconstruction error and feature sensitivity. DUO improves average AUROC by approximately 6% points over the strongest baselines while substantially reducing inference cost, and achieves SOTA on near-OOD and high-resolution benchmarks.

## FADE: Fractional Anomalous Dynamics Extrapolation for Training-Free Diffusion Transformers Acceleration

DevFeed: [FADE: Fractional Anomalous Dynamics Extrapolation for Training-Free Diffusion Transformers Acceleration](<https://devfeed.tech/articles/fade-fractional-anomalous-dynamics-extrapolation-for-training-free-diffusion-transformers-acceleration-125208.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10514>)

Author: Jinlong Yang, Lizilin, Jinke Wu, Yao Zhou

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusion-transformers](<https://devfeed.tech/topics/diffusion-transformers.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [acceleration](<https://devfeed.tech/tags/acceleration.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [caching](<https://devfeed.tech/tags/caching.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-transformers](<https://devfeed.tech/tags/diffusion-transformers.md>), [forecasting](<https://devfeed.tech/tags/forecasting.md>), [generative](<https://devfeed.tech/tags/generative.md>), [inference](<https://devfeed.tech/tags/inference.md>), [latency](<https://devfeed.tech/tags/latency.md>), [lut](<https://devfeed.tech/tags/lut.md>), [profiling](<https://devfeed.tech/tags/profiling.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [reuse](<https://devfeed.tech/tags/reuse.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [transformers](<https://devfeed.tech/tags/transformers.md>)

### AI overview

FADE is a training-free method intended to reduce iterative sampling costs in Diffusion Transformers. It models feature evolution as anomalous diffusion and uses a Mittag-Leffler fractional operator to adapt forecasts to curved, nonlinear latent trajectories, which the abstract says static integer-order methods handle poorly and may compound into errors. Offline profiling builds a multidimensional lookup table; during inference it retrieves parameters for spatiotemporal adaptation with reportedly negligible retrieval overhead. The abstract reports strong fidelity at aggressive skip intervals across architectures and modalities, but provides no specific benchmark figures in the supplied excerpt.

### Source excerpt

arXiv:2611.10514v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) have established themselves as the preeminent paradigm for high-fidelity generative modeling across various modalities, yet their practical deployment is hindered by the severe computational overhead of iterative sampling. While recent feature-caching methods mitigate this latency through feature reuse or forecasting based on integer-order operators, they remain limited in adapting to the non-stationary and highly curved dynamics of DiT latent trajectories. In this work, we reveal that DiT feature evolution exhibits anomalous diffusion characteristics and a certain degree of spatiotemporal heterogeneity, rendering static, integer-order forecasting operators inadequate and prone to error accumulation. To address this, we propose Fractional Anomalous Dynamics Extrapolation (FADE), a unified framework for training-free diffusion transformers acceleration. FADE models DiT feature evolution via anomalous diffusion and employs a Mittag--Leffler fractional operator to better adapt to high-curvature and non-linear latent trajectories of DiTs. Leveraging the empirical observation that these complex dynamics maintain cross-sample stability, we further construct a Look-Up Table with multiple dimensions through offline profiling. This enables negligible LUT-retrieval overhead, spatiotemporally adaptive parameter retrieval during online inference. Extensive evaluations across diverse architectures and modalities demonstrate that FADE achieves state-of-the-art acceleration, preserving exceptional generation fidelity under aggressive skip intervals with negligible latency overhead. Our project page is available at https://youngkinlon.github.io/fade-project-page/.

## Flex-DDCM: Zero-Shot Task-Oriented Image Compression via Diffusion Trajectory Guidance

DevFeed: [Flex-DDCM: Zero-Shot Task-Oriented Image Compression via Diffusion Trajectory Guidance](<https://devfeed.tech/articles/flex-ddcm-zero-shot-task-oriented-image-compression-via-diffusion-trajectory-guidance-125377.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11427>)

Author: Yixin Gao, Runsen Feng, Xin Li, Zhibo Chen

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Compression](<https://devfeed.tech/topics/compression.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [compression](<https://devfeed.tech/tags/compression.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [guidance](<https://devfeed.tech/tags/guidance.md>), [image](<https://devfeed.tech/tags/image.md>), [performance](<https://devfeed.tech/tags/performance.md>), [quality](<https://devfeed.tech/tags/quality.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

Flex-DDCM adapts zero-shot image compression for human viewing and machine analysis by selecting task-guided noise to modify Turbo-DDCM's coded diffusion trajectory. Guidance terms can be used separately or combined, with strengths that vary by denoising timestep; their effects are embedded in the compressed representation without adding decoding complexity. The abstract reports flexible trade-offs among pixel fidelity, perceptual quality, and downstream task performance, competitive results against trained task-oriented codecs, and faster decoding than Diff-ICMH.

### Source excerpt

arXiv:2611.11427v1 Announce Type: cross Abstract: Recent zero-shot image compression methods exploit the powerful priors of pretrained diffusion models to produce perceptually pleasing reconstructions without compression-specific training. However, image compression serves not only human viewing but also machine analysis, while existing zero-shot codecs have limited flexibility in meeting the diverse reconstruction requirements of these applications. In this paper, we propose Flex-DDCM, a framework for zero-shot task-oriented image compression via diffusion trajectory guidance. Building on Turbo-DDCM, we adapt the coded diffusion trajectory through task-guided noise selection, embedding the effects of task guidance into the compressed representation. Guidance terms can be applied individually or combined, with timestep-dependent strengths controlling their influence throughout denoising. This design enables task adaptation without introducing additional decoding complexity. Experiments demonstrate flexible trade-offs among pixel fidelity, perceptual quality, and downstream task performance, with competitive performance against trained task-oriented codecs and faster decoding than Diff-ICMH.

## Contrastive Distribution Matching for Amortized Sequential Monte Carlo in Discrete Diffusion

DevFeed: [Contrastive Distribution Matching for Amortized Sequential Monte Carlo in Discrete Diffusion](<https://devfeed.tech/articles/contrastive-distribution-matching-for-amortized-sequential-monte-carlo-in-discrete-diffusion-123164.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.23346>)

Author: Jaihoon Kim, Taehoon Yoon, Prin Phunyaphibarn, Seungjun Kim, Morteza Mardani, Minhyuk Sung

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI Development](<https://devfeed.tech/topics/ai-development.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [applications](<https://devfeed.tech/tags/applications.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [dna](<https://devfeed.tech/tags/dna.md>), [estimator](<https://devfeed.tech/tags/estimator.md>), [inference](<https://devfeed.tech/tags/inference.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [learning](<https://devfeed.tech/tags/learning.md>), [matching](<https://devfeed.tech/tags/matching.md>), [model](<https://devfeed.tech/tags/model.md>), [model-alignment](<https://devfeed.tech/tags/model-alignment.md>), [models](<https://devfeed.tech/tags/models.md>), [monte-carlo](<https://devfeed.tech/tags/monte-carlo.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [sequence](<https://devfeed.tech/tags/sequence.md>)

### AI overview

Contrastive Distribution Matching (CDM) reduces the inference burden of reward-tilted sampling with discrete diffusion by learning a parameterized twist function from positive and negative samples, amortizing part of Twisted Sequential Monte Carlo's costly twist estimation. Its training gradient uses the diffusion model's closed-form forward kernels. The abstract reports that evaluating the learned twist adds less than 5% overhead relative to one base-model forward pass and that CDM outperforms existing baselines at matched wall-clock time. Evaluations cover toxic text generation, regulatory DNA design, protein designability, and diffusion language model alignment; the supplied text does not provide detailed results for these tasks.

### Source excerpt

arXiv:2605.23346v2 Announce Type: replace Abstract: Discrete diffusion models have emerged as powerful frameworks for generating structured categorical data. However, efficiently sampling from reward-tilted distributions remains a fundamental challenge. While Twisted Sequential Monte Carlo (SMC) offers asymptotic exactness for this task, estimating the optimal twist function in discrete state spaces necessitates costly Monte Carlo approximations, resulting a severe computational bottleneck at inference. To overcome this limitation, we introduce Contrastive Distribution Matching (CDM), a novel framework that amortizes the cost of SMC inference by learning a parameterized twist function via positive and negative samples. For efficient training, we reformulate the gradient estimator to leverage the closed-form forward kernels of discrete diffusion models. In practice, evaluating our learned twist function incurs less than 5% additional computational overhead compared to a single forward pass of the base model. Through extensive empirical evaluations, we demonstrate that CDM consistently outperforms existing baselines under matched wall-clock time. We validate the effectiveness and versatility of our approach across a diverse range of applications, including toxic text generation, regulatory DNA sequence design, protein designability, and diffusion large language model alignment.

## Local Intrinsic Dimension Unveils Hallucinations in Diffusion Models

DevFeed: [Local Intrinsic Dimension Unveils Hallucinations in Diffusion Models](<https://devfeed.tech/articles/local-intrinsic-dimension-unveils-hallucinations-in-diffusion-models-124402.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.05026>)

Author: Bartlomiej Sobieski, Matthew Tivnan, Micha{\\l} Jan W{\\l}odarczyk, Dawid P{\\l}udowski, Pengfei Jin, Przemyslaw Biecek, Quanzheng Li

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Hallucination detection](<https://devfeed.tech/topics/hallucination-detection.md>), [Medical imaging](<https://devfeed.tech/topics/medical-imaging.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [consistency](<https://devfeed.tech/tags/consistency.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [filter](<https://devfeed.tech/tags/filter.md>), [hallucinations](<https://devfeed.tech/tags/hallucinations.md>), [imaging](<https://devfeed.tech/tags/imaging.md>), [intrinsic](<https://devfeed.tech/tags/intrinsic.md>), [local](<https://devfeed.tech/tags/local.md>), [medical](<https://devfeed.tech/tags/medical.md>), [medical-imaging](<https://devfeed.tech/tags/medical-imaging.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reduction](<https://devfeed.tech/tags/reduction.md>)

### AI overview

The paper frames structural hallucinations in diffusion models as instabilities on the model-induced manifold and identifies local intrinsic dimension (LID) as their primary driver. Its Intrinsic Quenching (IQ) mechanism reduces LID to alleviate hallucinations. The authors report that an instability-based filter matches or exceeds a recently proposed temporal filter, and that IQ outperforms standard reduction baselines across a range of benchmarks, with potential for improving anatomical consistency in medical imaging. The supplied text gives no benchmark details or implementation steps.

### Source excerpt

arXiv:2605.05026v2 Announce Type: replace-cross Abstract: Diffusion models are prone to generating structural hallucinations - samples that match the statistical properties of the training data yet defy underlying structural rules, resulting in anomalies like hands with more than five fingers. Recent research studied this failure mode from several viewpoints, offering partial explanations to their occurrence, such as mode interpolation. In this work, we propose a complementary perspective that treats hallucinations as instabilities on the model-induced manifold. We begin by showing that a hallucination filter based on such instabilities matches or exceeds the performance of the recently proposed temporal one. By tracing the source of these instabilities, we identify local intrinsic dimension (LID) as their primary driver and propose Intrinsic Quenching (IQ), a direct corrective mechanism that deflates it to alleviate hallucinations. IQ consistently outperforms standard hallucination reduction baselines across a wide array of benchmarks and offers a highly promising solution for enforcing anatomical consistency in downstream medical imaging tasks.

## FastWave: Optimized Diffusion Model for Audio Super-Resolution

DevFeed: [FastWave: Optimized Diffusion Model for Audio Super-Resolution](<https://devfeed.tech/articles/fastwave-optimized-diffusion-model-for-audio-super-resolution-120393.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.04122>)

Author: Nikita Kuznetsov, Maksim Kaledin

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [computational complexity](<https://devfeed.tech/topics/computational-complexity.md>), [Generative Adversarial Network](<https://devfeed.tech/topics/generative-adversarial-network.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [computational-complexity](<https://devfeed.tech/tags/computational-complexity.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-sd](<https://devfeed.tech/tags/cs-sd.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [flow](<https://devfeed.tech/tags/flow.md>), [inference](<https://devfeed.tech/tags/inference.md>), [model](<https://devfeed.tech/tags/model.md>), [optimized](<https://devfeed.tech/tags/optimized.md>), [super-resolution](<https://devfeed.tech/tags/super-resolution.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

FastWave is a compact diffusion model for audio super-resolution that estimates audio at 48 kHz from input at any sample rate. The abstract reports better results than NU-Wave 2 and performance comparable to state-of-the-art models. It has about 1.3 million parameters and 50 GFLOPs of computational complexity, and the authors report training it with fewer resources and significantly faster than most recent diffusion- and flow-based approaches. The abstract says its code is publicly available.

### Source excerpt

arXiv:2603.04122v2 Announce Type: replace-cross Abstract: Audio Super-Resolution is a set of techniques aimed at high-quality estimation of the given signal as if it would be sampled with higher sample rate. Among suggested methods there are diffusion and flow models (which are considered slower), generative adversarial networks (which are considered faster), however both approaches are currently presented by high-parametric networks, requiring high computational costs both for training and inference. We propose a solution to both these problems by re-considering the recent advances in the training of diffusion models and applying them to super-resolution from any to 48 kHz sample rate. Our approach shows better results than NU-Wave 2 and is comparable to state-of-the-art models. Our model called FastWave has around 50 GFLOPs of computational complexity and 1.3 M parameters and can be trained with less resources and significantly faster than the majority of recently proposed diffusion- and flow-based solutions. The code has been made publicly available.

## Flux as a Codec: Zero-Shot Lossy Compression with Flux

DevFeed: [Flux as a Codec: Zero-Shot Lossy Compression with Flux](<https://devfeed.tech/articles/flux-as-a-codec-zero-shot-lossy-compression-with-flux-120680.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07929>)

Author: Fuma Kimishima, Jinjia Zhou

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Compression](<https://devfeed.tech/topics/compression.md>), [flux](<https://devfeed.tech/topics/flux.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [codec](<https://devfeed.tech/tags/codec.md>), [coding](<https://devfeed.tech/tags/coding.md>), [compression](<https://devfeed.tech/tags/compression.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [flux](<https://devfeed.tech/tags/flux.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [image](<https://devfeed.tech/tags/image.md>), [image-compression](<https://devfeed.tech/tags/image-compression.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [neural](<https://devfeed.tech/tags/neural.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

FluxC uses Flux as a pretrained compression prior to perform zero-shot lossy coding. At matched SNR, the choice of coding path leaves the ideal information rate unchanged but alters how bits are allocated across time steps. The paper compares ancestral sampling with one-step and many-step ODE reconstruction; it reports lower distortion from ODE methods, particularly at low rates. Experiments on Gaussian sources and natural images support this result, and FluxC is reported to perform competitively with, and sometimes outperform, recent neural image compression methods. These claims are based on the abstract; it provides no quantitative results or implementation details.

### Source excerpt

arXiv:2611.07929v1 Announce Type: new Abstract: We propose FluxC, a zero-shot codec that uses Flux as a compression prior. Although DiffC has shown that pretrained diffusion models can serve as zero-shot codecs, we ask how the choice of coding path and reconstruction method affects rate--distortion--perception behavior. We analyze FluxC along three axes: coding cost and bit allocation, reconstruction method, and low rate distortion. At matched signal-to-noise ratio (SNR), we show that the path choice does not affect the ideal information rate, but changes how bits are allocated across time steps. For reconstruction, we compare ancestral sampling, one-step ODE reconstruction, and many-step ODE reconstruction. Our analysis shows that ODE-based reconstruction can achieve lower distortion than ancestral sampling, especially in the low rate regime. Experiments on Gaussian sources and natural images support this prediction. Moreover, FluxC performs competitively with, and in some cases outperforms, recent neural image compression methods, suggesting that Flux can serve as a principled zero-shot codec.

## BridgeDock: Taming Generative Co-Folding Prior for Molecular Docking with Diffusion Bridge

DevFeed: [BridgeDock: Taming Generative Co-Folding Prior for Molecular Docking with Diffusion Bridge](<https://devfeed.tech/articles/bridgedock-taming-generative-co-folding-prior-for-molecular-docking-with-diffusion-bridge-115093.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.00795>)

Author: Shikun Feng, Zipeng Yang, Qiuyi Li, Mingze Yin, Xiaoyuan Zhang, Tingjun Hou, Yiheng Zhu

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [alphafold](<https://devfeed.tech/tags/alphafold.md>), [alphafold-3](<https://devfeed.tech/tags/alphafold-3.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [flexible](<https://devfeed.tech/tags/flexible.md>), [folding](<https://devfeed.tech/tags/folding.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [generate](<https://devfeed.tech/tags/generate.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [inference](<https://devfeed.tech/tags/inference.md>), [input](<https://devfeed.tech/tags/input.md>), [model](<https://devfeed.tech/tags/model.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [models](<https://devfeed.tech/tags/models.md>), [noise](<https://devfeed.tech/tags/noise.md>)

### AI overview

BridgeDock adapts pretrained co-folding models for flexible protein-ligand docking by treating the apo-to-holo change as a diffusion bridge. It generates from the initial protein-ligand state instead of Gaussian noise, and aligns bridge states with the pretrained denoiser's original schedule. On standard flexible-docking benchmarks, it reportedly outperforms strong co-folding baselines, with efficient inference and effectiveness even at very few denoising steps. The excerpt does not provide benchmark results or implementation details.

### Source excerpt

arXiv:2611.00795v1 Announce Type: new Abstract: Molecular docking aims to predict the three-dimensional structure of a protein-ligand complex; in the conventional structure-based setting, it is conditioned on an input protein structure and ligand, often starting from an apo receptor in flexible docking or a holo receptor in rigid docking. Recent co-folding models exemplified by AlphaFold 3 achieve strong accuracy in protein-ligand complex prediction, but they typically generate complexes from sequence-derived inputs and therefore do not directly use apo structures as structural priors in docking, leading to a larger conformational search space and weaker robustness in difficult or few-step settings. We present BridgeDock, a framework that adapts pretrained co-folding backbones to flexible docking by modeling the apo-to-holo transition as a diffusion bridge, allowing generation from the initial protein-ligand state rather than from Gaussian noise. To make this bridge formulation compatible with pretrained denoisers, we introduce an alignment mechanism that maps bridge states to the original denoising schedule of the pretrained model. Experiments on standard flexible docking benchmarks show that BridgeDock consistently outperforms strong co-folding baselines. Moreover, BridgeDock is computationally efficient at inference time and remains effective even with very few denoising steps.

## Cumulative Marginal Matching for One-Step Diffusion

DevFeed: [Cumulative Marginal Matching for One-Step Diffusion](<https://devfeed.tech/articles/cumulative-marginal-matching-for-one-step-diffusion-111468.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.34575>)

Author: Paribesh Regmi, Rui Li

Published: 2026-11-02T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [approximation](<https://devfeed.tech/tags/approximation.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [distributions](<https://devfeed.tech/tags/distributions.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [generation](<https://devfeed.tech/tags/generation.md>), [inference](<https://devfeed.tech/tags/inference.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [models](<https://devfeed.tech/tags/models.md>), [quality](<https://devfeed.tech/tags/quality.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Cumulative Marginal Matching (CMM) trains few-step flow-matching models to reduce cumulative Wasserstein-2 discrepancies between the base and few-step models' intermediate marginal distributions across the generation sequence. The authors turn the otherwise intractable objective into a tractable surrogate using samples from intermediate marginals generated by conditional interpolation. They also show that Shortcut-CSL is a first-order approximation of CMM. On standard benchmarks, CMM reportedly improves generation quality over Shortcut-CSL and other one- and few-step baselines under identical training budgets.

### Source excerpt

arXiv:2610.34575v2 Announce Type: replace Abstract: The sampling procedures of diffusion and flow-matching models involve hundreds of denoising steps incurring substantial inference time, which has motivated a growing body of work on training schemes for few-step and one-step generation. Existing approaches primarily seek to align the generation trajectory of few-step and base models, yet their training objectives only involve discrepancies at a single point in the trajectory, encouraging only local \emph{point-level} alignment. While recent methods such as \emph{Shortcut-CSL} introduce cumulative objective that enforce alignment over the entire trajectory, this is suboptimal as they do not directly enforce alignment between the distributions induced by the two models throughout the generation process. We propose \emph{Cumulative Marginal Matching} (CMM), a principled objective for few-step generation: by viewing flow-matching as a sequence of intermediate marginal distributions connecting the prior and data distributions, CMM directly optimizes the cumulative discrepancy between the base and few-step marginal distributions over the entire sequence. Specifically, we formulate this discrepancy using the Wasserstein-2 distance. By exploiting the pushforward representation of flow-matching, we transform an intractable cumulative Wasserstein objective into a tractable surrogate that can be optimized using samples from the intermediate marginal, which are readily generated through the conditional interpolation process of flow-matching. Furthermore, we show that the existing \emph{Shortcut-CSL} objective arises as a first-order approximation of that of the CMM, establishing a theoretical connection between the two objectives. Experiments on standard benchmarks demonstrate improvement in generation quality over \emph{Shortcut-CSL} and other state-of-the-art one- and few-step baselines under identical training budgets.

## D-MOD: Diffusion mode estimation for accelerated MRI

DevFeed: [D-MOD: Diffusion mode estimation for accelerated MRI](<https://devfeed.tech/articles/d-mod-diffusion-mode-estimation-for-accelerated-mri-108872.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.40110>)

Author: Md Kawsar Ahmed, Mathews Jacob

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [accelerated](<https://devfeed.tech/tags/accelerated.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [approach](<https://devfeed.tech/tags/approach.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [brain](<https://devfeed.tech/tags/brain.md>), [but](<https://devfeed.tech/tags/but.md>), [combines](<https://devfeed.tech/tags/combines.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [fast](<https://devfeed.tech/tags/fast.md>), [faster](<https://devfeed.tech/tags/faster.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [instead](<https://devfeed.tech/tags/instead.md>), [iteration](<https://devfeed.tech/tags/iteration.md>), [lower-bound](<https://devfeed.tech/tags/lower-bound.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [noise](<https://devfeed.tech/tags/noise.md>)

### AI overview

D-MOD produces a deterministic posterior-mode reconstruction for accelerated MRI, avoiding the need to average many posterior samples. At each iteration, it draws noisy samples of the current estimate, denoises them with a pretrained diffusion model, then combines their average with measured k-space data through a conjugate-gradient solve. A minorize-maximize procedure on a lower bound of the smoothed log-posterior provably increases the objective, while noise levels are annealed. On fastMRI brain T1, T2, and FLAIR data, one D-MOD reconstruction exceeded the PSNR of 10-sample finite-MMSE estimates from DPS, DAPS, and SNORE-P, and came within 0.31 dB of DiffPIR and DDS. The abstract reports D-MOD was 11-73 times faster than those MMSE estimates.

### Source excerpt

arXiv:2610.40110v1 Announce Type: new Abstract: Diffusion models are powerful priors for accelerated MRI reconstruction, but most current methods are posterior samplers, which return a random draw from the posterior. A reproducible and high-quality reconstruction often requires averaging many samples, resulting in a high computational cost. We introduce D-MOD, a fast algorithm that instead provides a posterior mode. It uses a minorize-maximize algorithm on a lower bound of the smoothed log-posterior, which provably increases the objective. Each iteration draws S noisy samples of the current estimate at the current noise level, denoises them with a pretrained diffusion model, and combines their average with the measured k-space through a conjugate-gradient (CG) solve. We anneal the noise levels, as the distance between the bound and its target decays quadratically with noise. On fastMRI brain (T1, T2, FLAIR), a single D-MOD reconstruction exceeds, in PSNR, the 10-sample finite-MMSE estimates of DPS, DAPS and SNORE-P and stays within 0.31 dB of the strongest baselines (DiffPIR, DDS). This approach thus provides a fast deterministic reconstruction algorithm (11-73 times faster than MMSE estimates), which is useful in a clinical setting.

## Composing and Steering Diffusion Models with Generalized Feynman-Kac Correctors

DevFeed: [Composing and Steering Diffusion Models with Generalized Feynman-Kac Correctors](<https://devfeed.tech/articles/composing-and-steering-diffusion-models-with-generalized-feynman-kac-correctors-108901.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.40383>)

Author: Wael Mattar, Daniel Freedman

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [generative](<https://devfeed.tech/tags/generative.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [reward](<https://devfeed.tech/tags/reward.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [steering](<https://devfeed.tech/tags/steering.md>)

### AI overview

Generalized Feynman-Kac correctors (GFKC) provide a way to sample from arbitrary target functions combining multiple pretrained diffusion models, including targets tilted by a downstream reward. The method derives a closed-form Feynman-Kac potential and tracks each model's density along trajectories driven by another velocity field, correcting samples during generation in the log domain without retraining or fine-tuning. The authors report close agreement with ground-truth targets on analytic Gaussian models and non-trivial model combinations and reward steering on diffusion models trained on MNIST.

### Source excerpt

arXiv:2610.40383v1 Announce Type: new Abstract: Diffusion models have become the leading paradigm in generative modeling, yet many applications require sampling from distributions that are not those of any single pretrained model, but are instead defined through combinations of several models, possibly tilted toward high values of a downstream reward. Naively combining the scores of the models yields incorrect samples, while existing corrections are restricted to specific families of targets. In this work, we introduce the generalized Feynman-Kac correctors (GFKC), a principled framework for sampling from an arbitrary target function of several pretrained diffusion models. We show that the geometric interpolation between an arbitrary initial function and the target function of the models satisfies a Feynman--Kac partial differential equation, and derive its potential in closed form. To evaluate this potential during generation, we further generalize the log-probability ODE to track the density of each model along trajectories driven by a different velocity field. The resulting algorithm operates entirely in the log domain, incorporates reward-based steering at no additional conceptual cost, and corrects the generation on the fly, without retraining or fine-tuning the models. We validate the method on analytic Gaussian models, where it closely matches the ground-truth targets, and on diffusion models trained on MNIST, where it samples from non-trivial combinations of the models and enables effective steering.

## Improved Sampling Schedules for Discrete Diffusion Models

DevFeed: [Improved Sampling Schedules for Discrete Diffusion Models](<https://devfeed.tech/articles/improved-sampling-schedules-for-discrete-diffusion-models-108465.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.06849>)

Author: Alberto Foresti, Mustapha Bounoua, Giulio Franzese, Luca Ambrogioni, Pietro Michiardi

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [language](<https://devfeed.tech/tags/language.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [mbpp](<https://devfeed.tech/tags/mbpp.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [schedules](<https://devfeed.tech/tags/schedules.md>)

### AI overview

The paper proposes two sampling schedules for uniform discrete diffusion models, derived by analyzing reverse-process dynamics. Entropic Discrete Schedule (EDS) aligns with entropy production; Wasserstein Discrete Schedule (WDS) combines entropy production and dynamical activity as a geometric cost. In language modeling experiments, WDS reportedly matches uniform scheduling on the MBPP benchmark while using up to half the computations. The supplied abstract does not specify further experimental settings or results.

### Source excerpt

arXiv:2602.06849v3 Announce Type: replace Abstract: Uniform Discrete Diffusion Models have emerged as a powerful paradigm for generative modeling on sequence data; however, the information-theoretic principles governing their reverse processes remain underexplored. In this work, we bridge this gap by analyzing the reverse process dynamics through the lens of thermodynamic entropy production and dynamical activity. We use a variational expression of the discrete Wasserstein distance to characterize the dynamics of the generation process. Leveraging these insights, we first introduce the Entropic Discrete Schedule (EDS), which aligns with entropy production, and later construct the Wasserstein Discrete Schedule (WDS), which combines entropy production and dynamical activity to act as a geometric cost. Our experimental results on language modeling show that WDS provide the same results as uniform while using up to half of the computations on the MBPP benchmark. The code is available at https://github.com/AlbertoForesti/impscheds.

## Diffusion models for the $O(3)$ nonlinear sigma model

DevFeed: [Diffusion models for the $O(3)$ nonlinear sigma model](<https://devfeed.tech/articles/diffusion-models-for-the-o-3-nonlinear-sigma-model-108822.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.39568>)

Author: Daniel Schubring

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [approximation](<https://devfeed.tech/tags/approximation.md>), [cond-mat-stat-mech](<https://devfeed.tech/tags/cond-mat-stat-mech.md>), [correlation](<https://devfeed.tech/tags/correlation.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diagnostics](<https://devfeed.tech/tags/diagnostics.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [hep-lat](<https://devfeed.tech/tags/hep-lat.md>), [neural-networks](<https://devfeed.tech/tags/neural-networks.md>), [sampling](<https://devfeed.tech/tags/sampling.md>)

### AI overview

A score-based diffusion model is trained on the two-dimensional classical Heisenberg model, a lattice representation of the O(3) nonlinear sigma model. Training networks on three coarse-grainings of the same data enables multiscale sampling, which modestly improves accuracy and computational cost when sampling directly on the finest lattice. The work also introduces Stein (Schwinger-Dyson) diagnostics for the two-point correlation function to assess the learned score without reverse diffusion; these diagnostics are added to a training loss. Implementation considerations for spins on S² include lifting network inputs to S³ in the CP¹ representation and using a boundary-layer approximation to the short-time spherical heat kernel to extend its usefulness to larger diffusion times.

### Source excerpt

arXiv:2610.39568v1 Announce Type: cross Abstract: A score-based diffusion model is trained on the two-dimensional classical Heisenberg model, the lattice representation of the $O(3)$ nonlinear sigma model. Neural networks are trained on three different coarse-grainings of the same training set to implement a multiscale sampling scheme that modestly improves both the accuracy and computational cost of direct sampling on the finest lattice. Stein (or Schwinger-Dyson) diagnostics for the two-point correlation function are introduced to assess the quality of the learned score without relying on reverse diffusion sampling. These diagnostics are also incorporated into an additional training loss. We discuss several technical considerations relevant to implementing diffusion models on $S^2$. These include a consideration of lifting the network input to $S^3$, as in the $CP^1$ representation, and developing a boundary-layer approximation to the short-time spherical heat kernel that extends its usefulness to larger diffusion times.

## Adapting Video Models for Robot Control via Motion-guided Asymmetric Prediction

DevFeed: [Adapting Video Models for Robot Control via Motion-guided Asymmetric Prediction](<https://devfeed.tech/articles/adapting-video-models-for-robot-control-via-motion-guided-asymmetric-prediction-103165.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.36624>)

Author: Taeyoung Kim, Junyoung Sung, Heeseung Kwon, Minseop Kim, Jinwoo Shin

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [Robot Manipulation](<https://devfeed.tech/topics/robot-manipulation.md>), [Model Development](<https://devfeed.tech/topics/model-development.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [adapting](<https://devfeed.tech/tags/adapting.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [models](<https://devfeed.tech/tags/models.md>), [motion](<https://devfeed.tech/tags/motion.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [robot](<https://devfeed.tech/tags/robot.md>), [video](<https://devfeed.tech/tags/video.md>)

### AI overview

MoGAP improves robot policy performance by directing video prediction supervision toward moving regions. It adds an auxiliary training path to the shared video diffusion backbone: dynamic regions receive more corruption and must be reconstructed using less-corrupted static regions as context. Motion cues come from training videos, with no external teacher or additional modalities required. Across manipulation benchmarks and WAM architectures, the abstract reports higher success on the GR-1 tabletop benchmark: Fast-WAM (Wan2.2-5B) rises from 38.8% to 47.5%, and Cosmos3-Nano from 52.3% to 62.3%. The supplied text is an abstract, so it does not provide further experimental details.

### Source excerpt

arXiv:2610.36624v1 Announce Type: cross Abstract: World Action Models (WAMs) leverage pretrained video generative models for robotic control by jointly predicting future visual observations and actions. The visual prediction objective plays a key role in adapting pretrained video representations to the target robot domain. Yet standard visual prediction inherited from video pretraining treats all visual regions uniformly, despite their differing relevance to robotic control. We therefore investigate the spatial allocation of visual supervision in WAMs and show that focusing supervision on dynamic regions can substantially improve policy performance. Building on this finding, we introduce Motion-Guided Asymmetric Prediction (MoGAP), which augments WAM training with motion-focused visual prediction. MoGAP allocates noise asymmetrically across regions, training the model to reconstruct more heavily corrupted dynamic regions using less corrupted static regions as context. MoGAP is trained alongside standard video-action prediction through a training-only auxiliary prediction path within the shared video diffusion backbone. It uses motion cues derived directly from training videos, requiring neither an external teacher nor additional modalities. Across robot manipulation benchmarks and WAM architectures, MoGAP consistently improves policy performance, increasing success rates on the GR-1 tabletop benchmark from 38.8% to 47.5% for Fast-WAM (Wan2.2-5B) and from 52.3% to 62.3% for Cosmos3-Nano.

## PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models

DevFeed: [PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models](<https://devfeed.tech/articles/phystrace-interpretable-temporal-evaluation-of-physical-consistency-in-video-generative-models-98218.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.31429>)

Author: Taekoan Yoo, Wonkyung Jung, Kyunghun Kim, Kyeongbo Kong

Published: 2026-10-23T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [Generative AI](<https://devfeed.tech/topics/generative-ai.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Video Analytics](<https://devfeed.tech/topics/video-analytics.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>)

Tags: [consistency](<https://devfeed.tech/tags/consistency.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [flow](<https://devfeed.tech/tags/flow.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [temporal](<https://devfeed.tech/tags/temporal.md>)

### AI overview

PhysTrace is a training-free framework that aligns MLLM-generated expectations conditioned on video history with observed windows to locate physical violations in time and describe them, without task-specific probes or classifiers. The paper also introduces IntPhys2-Timing, which adds violation intervals to IntPhys2. PhysTrace reports up to 70.7% hit rate across physical principles on IntPhys2-Timing and higher overall possible-versus-impossible discrimination accuracy than the compared methods on IntPhys2. Applied to reconstructed trajectories from pretrained diffusion and flow-based video models, it finds attribute- and noise-dependent patterns in where violations occur over time.

### Source excerpt

arXiv:2610.31429v1 Announce Type: new Abstract: World action models increasingly rely on pretrained video models assumed to capture physical dynamics for prediction, planning, and action generation. Existing evaluations assess plausibility, probe representations, or measure predictive surprise, but rarely identify which history-conditioned physical expectations conflict with observations. We introduce PhysTrace, a training-free framework aligning MLLM-generated, history-conditioned expectations with observed video windows to temporally localize physical violations and provide interpretable descriptions without task-specific probes or classifiers. We also introduce IntPhys2-Timing, adding violation intervals to IntPhys2. PhysTrace achieves a Hit Rate of up to 70.7% across physical principles on IntPhys2-Timing and outperforms the compared methods in overall possible-impossible discrimination accuracy on IntPhys2. Applied to reconstructed videos along the trajectories of pretrained diffusion/flow-based video models, PhysTrace reveals attribute- and noise-dependent patterns in physical violation localization, enabling interpretable physical consistency analysis across timesteps.

## PhysibleMDM: Enforcing Inference-Time Physical Feasibility Constraints on Motion Diffusion Models via Noise Optimization

DevFeed: [PhysibleMDM: Enforcing Inference-Time Physical Feasibility Constraints on Motion Diffusion Models via Noise Optimization](<https://devfeed.tech/articles/physiblemdm-enforcing-inference-time-physical-feasibility-constraints-on-motion-diffusion-models-via-noise-optimization-98072.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.30500>)

Author: Maximilian St\\"olzle, David M\\"uller, Agon Serifi, Ruben Grandia, Espen Knoop, Moritz B\\"acher

Published: 2026-10-23T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Robotics](<https://devfeed.tech/topics/robotics.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Operations research and optimization](<https://devfeed.tech/topics/operations-research-and-optimization.md>)

Tags: [acceleration](<https://devfeed.tech/tags/acceleration.md>), [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [bipedal-robot](<https://devfeed.tech/tags/bipedal-robot.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [es](<https://devfeed.tech/tags/es.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [humanoid](<https://devfeed.tech/tags/humanoid.md>), [inference](<https://devfeed.tech/tags/inference.md>), [mdm](<https://devfeed.tech/tags/mdm.md>), [motion](<https://devfeed.tech/tags/motion.md>), [noise](<https://devfeed.tech/tags/noise.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [penetration](<https://devfeed.tech/tags/penetration.md>), [program](<https://devfeed.tech/tags/program.md>), [robot](<https://devfeed.tech/tags/robot.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [strategy](<https://devfeed.tech/tags/strategy.md>)

### AI overview

PhysibleMDM improves the physical feasibility of motion diffusion model output by optimizing diffusion noise at inference time, without retraining the generator. It evaluates surrogate constraints for ground penetration, slip, self-collision and dynamics, with an inverse-dynamics inner loop formulated as a cone program. Sep-CMA-ES solves the resulting black-box optimization without relying on costly, discontinuous gradients, and an autoregressive scheme chains optimized segments into longer sequences. Across 100 motion styles on an SMPL humanoid, it reduced mean ground penetration by 75%, self-collision penetration by 92% and ground slip by 74% relative to unoptimized output. In 11 custom-biped sequences, median reductions were 89.5% for per-sequence mean acceleration-reconstruction error and 79.5% for mean inverse-dynamics residual.

### Source excerpt

arXiv:2610.30500v1 Announce Type: cross Abstract: Motion Diffusion Models (MDMs) can generate diverse and visually appealing human motion; yet, the resulting trajectories frequently violate physical constraints, limiting their downstream use in simulation and control. We present PhysibleMDM, a framework that promotes physical feasibility of MDM output by optimizing at inference time the diffusion noise without retraining the generative model. We introduce a set of surrogate feasibility constraints that can be evaluated without full trajectory simulation. These constraints combine kinematic checks (e.g., ground penetration, slip, and self-collision) with dynamic feasibility tests computed via an inverse-dynamics inner loop formulated as a cone program. We solve the resulting black-box noise optimization problem using the Separable Covariance Matrix Adaptation Evolution Strategy (Sep-CMA-ES), thereby avoiding reliance on computationally expensive and discontinuous gradients. An autoregressive scheme then chains optimized segments to synthesize long motion sequences. Across 100 motion styles on an SMPL humanoid, PhysibleMDM reduces mean ground penetration, self-collision penetration, and ground slip by 75%, 92%, and 74%, respectively, relative to unoptimized diffusion output. Across 11 sequences for a custom bipedal robot, it achieves median reductions of 89.5% in per-sequence mean acceleration-reconstruction error and 79.5% in the mean inverse-dynamics residual.

## How to Code Stable Diffusion Directly in Python on Runpod

DevFeed: [How to Code Stable Diffusion Directly in Python on Runpod](<https://devfeed.tech/articles/how-to-code-stable-diffusion-directly-in-python-on-runpod-94591.md>)

Original publisher: [Read original article](<https://www.runpod.io/blog/stable-diffusion-python-runpod>)

Author: Brendan McKeag

Published: 2026-10-22T02:42:59Z

Content type: tutorial

Language: en

Sources: [Runpod Blog.](<https://devfeed.tech/sources/runpod-blog.md>)

Topics: [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [Jupyter Notebook](<https://devfeed.tech/topics/jupyter-notebook.md>), [Python](<https://devfeed.tech/topics/python.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [how-to](<https://devfeed.tech/tags/how-to.md>), [jupyter-notebook](<https://devfeed.tech/tags/jupyter-notebook.md>), [python](<https://devfeed.tech/tags/python.md>), [stable-diffusion](<https://devfeed.tech/tags/stable-diffusion.md>)

### AI overview

The excerpt explains how to generate Stable Diffusion images from Python in Jupyter Notebook on Runpod, without installing a front end. It directs readers to deploy the official Stable Diffusion template, connect to the pod on port 8888, and run notebook examples that download a Hugging Face model and generate images. It also describes an SDXL example with a refiner and a loop that reuses a seed to save images at different denoising steps, with Pillow used to combine them for comparison. The excerpt says this makes iterative comparisons easier than repeatedly adjusting a UI slider and starting generations. CivitAI models are another option, but downloading them requires a free API key.

### Source excerpt

Skip the front ends, learn how to use Jupyter Notebook on Runpod to run Stable Diffusion directly in Python. Great for devs who want full control.

## RECAP: Reciprocal Sparse Attention and Hierarchical Caching for Efficient Video Diffusion

DevFeed: [RECAP: Reciprocal Sparse Attention and Hierarchical Caching for Efficient Video Diffusion](<https://devfeed.tech/articles/recap-reciprocal-sparse-attention-and-hierarchical-caching-for-efficient-video-diffusion-93607.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.26244>)

Author: Tiange Xu, Kai Liu, Linghe Kong, Yulun Zhang

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [Cache](<https://devfeed.tech/topics/cache.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [caching](<https://devfeed.tech/tags/caching.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [generation](<https://devfeed.tech/tags/generation.md>), [recap](<https://devfeed.tech/tags/recap.md>), [video](<https://devfeed.tech/tags/video.md>)

### AI overview

RECAP combines sparse attention with temporal caching in video diffusion using reciprocal correction and online error feedback. Cached dense refinement corrects omitted interactions in current sparse estimates; fresh sparse statistics track input changes and regulate the aging correction. Corrected predictions then renew the sources used for whole-step reuse, allowing prediction history to advance between dense refreshes. The method is training-free. Across multiple video diffusion models, the authors compare it with sparse-attention methods, caching methods, and direct compositions. On HunyuanVideo at 720p for 5-second generation, RECAP reports a 3.05x speedup and better fidelity and inference speed than all tested direct compositions.

### Source excerpt

arXiv:2610.26244v1 Announce Type: cross Abstract: Diffusion transformers generate high-quality videos but remain computationally expensive. Sparse attention and temporal caching reduce computation within and across denoising steps, respectively, offering a natural composition. However, predictions produced with sparse attention become sources for temporal reuse, coupling their errors along the sampling trajectory. We observe that improving the accuracy of these sources recovers fidelity without changing the reuse schedule, and that sparsification errors remain directionally aligned across nearby steps, suggesting that correction information can be shared rather than recomputed. These observations motivate RECAP, a training-free method that couples sparse attention with hierarchical caching through reciprocal correction, adaptively coordinated by online error feedback. Cached dense refinement corrects omitted interactions in current sparse estimates, while fresh sparse statistics track input changes and regulate the aging correction. The resulting predictions then renew the sources for whole-step prediction reuse, allowing prediction history to advance between dense refreshes. We evaluate RECAP on multiple video diffusion models against sparse-attention methods, caching methods, and their direct compositions. On HunyuanVideo at 720p 5-second generation, RECAP achieves a 3.05x speedup and outperforms all tested direct compositions in both fidelity and inference speed.

## Learn from Your Mistakes: Self-Correcting Masked Diffusion Models

DevFeed: [Learn from Your Mistakes: Self-Correcting Masked Diffusion Models](<https://devfeed.tech/articles/learn-from-your-mistakes-self-correcting-masked-diffusion-models-94105.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.11590>)

Author: Yair Schiff, Omer Belhasin, Roy Uziel, Guanghan Wang, Marianne Arriola, Gilad Turok, Ran Zilberstein, Michael Elad, Volodymyr Kuleshov

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [diffusers](<https://devfeed.tech/topics/diffusers.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [error](<https://devfeed.tech/tags/error.md>), [faster](<https://devfeed.tech/tags/faster.md>), [generation](<https://devfeed.tech/tags/generation.md>), [inference](<https://devfeed.tech/tags/inference.md>), [mistakes](<https://devfeed.tech/tags/mistakes.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [performance](<https://devfeed.tech/tags/performance.md>), [quality](<https://devfeed.tech/tags/quality.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [token](<https://devfeed.tech/tags/token.md>), [train](<https://devfeed.tech/tags/train.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Masked diffusion models keep unmasked tokens fixed, so early errors can accumulate. Progressive Self-Correction (ProSeCo) trains a model to both unmask and correct tokens, reusing denoising-network outputs for corrector training; at generation time, it inserts refinement steps that can revise already decoded tokens. The abstract reports tests across conditional and unconditional tasks, with sampling up to about 4x faster than the best-accuracy baseline while still outperforming it, and benchmark gains up to about 1.2x when using slower, higher-NFE decoding.

### Source excerpt

arXiv:2602.11590v4 Announce Type: replace Abstract: Masked diffusion models (MDMs) have emerged as a promising alternative to autoregressive models, enabling parallel token generation while achieving competitive performance. Despite these advantages, MDMs face a fundamental limitation: once tokens are unmasked, they remain fixed, leading to error accumulation and ultimately degrading sample quality. We address this by proposing a framework that trains a model to perform both unmasking and correction. By reusing outputs from the MDM denoising network as inputs for corrector training, we train a model to recover from potential mistakes. During generation we apply additional corrective refinement steps between unmasking ones in order to change decoded tokens and improve outputs. We name our training and sampling method Progressive Self-Correction (ProSeCo) for its unique ability to iteratively refine an entire sequence, including already generated tokens. We conduct extensive experimental validation across multiple conditional and unconditional tasks, demonstrating that ProSeCo yields better quality-efficiency trade-offs (up to ~4x faster sampling while still outperforming the best accuracy baseline method) and enables further gains through inference-time compute scaling (up to ~1.2x improvement on benchmarks when allowing for slower, high-NFE decoding).

## Disco Diffusion on Runpod: Creative AI for Artists

DevFeed: [Disco Diffusion on Runpod: Creative AI for Artists](<https://devfeed.tech/articles/disco-diffusion-on-runpod-creative-ai-for-artists-88378.md>)

Original publisher: [Read original article](<https://www.runpod.io/blog/disco-diffusion-runpod>)

Author: Zhen Lu

Published: 2026-10-15T22:15:10Z

Content type: tutorial

Language: en

Sources: [Runpod Blog.](<https://devfeed.tech/sources/runpod-blog.md>)

Topics: [diffusers](<https://devfeed.tech/topics/diffusers.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [compute](<https://devfeed.tech/tags/compute.md>), [connect](<https://devfeed.tech/tags/connect.md>), [container](<https://devfeed.tech/tags/container.md>), [creative-ai](<https://devfeed.tech/tags/creative-ai.md>), [dashboard](<https://devfeed.tech/tags/dashboard.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [download](<https://devfeed.tech/tags/download.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-compute](<https://devfeed.tech/tags/gpu-compute.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [image](<https://devfeed.tech/tags/image.md>)

### AI overview

This tutorial explains how to run Disco Diffusion on Runpod, from choosing and deploying a GPU pod to connecting to the art interface and queuing image renders. It notes that GPU memory needs vary by settings and that spot instances may be interrupted.

### Source excerpt

Explore Disco Diffusion on Runpod, an experimental art model beloved for its dreamlike style. Perfect for creative pros looking to generate high-concept.

[Next page](<https://devfeed.tech/topics/diffusers.md?cursor=WyIyMDI2LTEwLTE1VDIyOjE1OjEwKzAwOjAwIiwgIjM5MjQxMTBlLWRmMTQtNDk3YS1hYmNiLTAxNTQ0ZWExN2U0OCJd>)