图像生成 - arXivDaily 专题

2506.07822 2026-02-09 cs.LG cs.AI 78%

Accelerating Diffusion Planners in Offline RL via Reward-Aware Consistency Trajectory Distillation

通过奖励感知一致性轨迹蒸馏加速离线强化学习中的扩散规划器

Xintong Duan, Yutong He, Fahim Tajwar, Ruslan Salakhutdinov, J. Zico Kolter, Jeff Schneider

机构 * Carnegie Mellon University（卡内基梅隆大学）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出了一种奖励感知一致性轨迹蒸馏方法，通过解耦训练和无噪声奖励信号，在离线强化学习中实现高效扩散规划器，提升性能并加快推理速度。

2509.21058 2026-02-05 cs.LG 78%

SPREAD: Sampling-based Pareto front Refinement via Efficient Adaptive Diffusion

SPREAD: 基于采样的帕累托前沿精炼 via 高效自适应扩散

Sedjro Salomon Hotegni, Sebastian Peitz

机构 * Department of Computer Science, TU Dortmund University（计算机科学系，多特蒙德技术大学）； Lamarr Institute for Machine Learning and Artificial Intelligence（拉马尔机器学习与人工智能研究所）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结 SPREAD通过高效自适应扩散方法提升多目标优化的帕累托前沿覆盖效率

2602.03627 2026-02-04 cs.LG 78%

Ultra Fast PDE Solving via Physics Guided Few-step Diffusion

通过物理引导的少步扩散实现超快PDE求解

Cindy Xiangrui Kong, Yueqi Wang, Haoyang Zheng, Weijian Luo, Guang Lin

机构 * Purdue University（普渡大学）； hi-Lab, Xiaohongshu Inc（小红书实验室，小红书公司）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结 Phys-Instruct通过物理引导的蒸馏框架，实现PDE求解的高效与高物理一致性。

2601.20706 2026-01-29 cs.AR cs.AI cs.DC 78%

Beyond GEMM-Centric NPUs: Enabling Efficient Diffusion LLM Sampling

超越以GEMM为中心的NPUs：使高效的扩散LLM采样成为可能

Binglei Lou, Haoran Wu, Yao Lai, Jiayi Nie, Can Xiao, Xuan Guo, Rika Antonova, Robert Mullins, Aaron Zhao

机构 * Imperial College London（伦敦帝国学院）； University of Edinburgh（爱丁堡大学）； University of Cambridge（剑桥大学）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出了一种针对扩散LLM采样的NPU架构优化方案，通过轻量级向量原语、内存重用策略和混合精度内存层次结构，实现了2.53倍的加速性能。

2601.15210 2026-01-28 physics.optics math-ph math.MP 78%

Enhanced posterior sampling via diffusion models for efficient metasurfaces inverse design

通过扩散模型增强后验采样以实现高效的超材料逆设计

Mathys Le Grand, Pascal Urard, Denis Rideau, Loumi Trémas, Damien Maitre, Louis-Henri Fernandez-Mouron, Adam Fuchs, Régis Orobtchouk

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出基于扩散模型的逆设计方法，通过增强后验采样和一致性约束，实现高效且高精度的超材料设计。

Comments 32 pages, 16 figures; Abstract typos corrected, errors corrected in Table 2 and Figure 6; Typos corrected

2601.01594 2026-01-26 stat.ML cs.LG 78%

Variance-Reduced Diffusion Sampling via Target Score Identity

通过目标分数身份实现方差缩减的扩散采样

Alois Duston, Tan Bui-Thanh

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出通过目标分数身份实现扩散采样中的方差缩减，结合非参数重要性采样和状态依赖的混合规则，提升样本质量。

Comments Updated to match journal submission and add ACM & MSC class info

2505.08142 2026-01-21 eess.IV 78%

Highly Undersampled MRI Reconstruction via a Single Posterior Sampling of Diffusion Models

通过扩散模型单步后采样实现高 undersampling 的 MRI 重建

Jin Liu, Qing Lin, Zhuang Xiong, Shanshan Shan, Chunyi Liu, Min Li, Feng Liu, G. Bruce Pike, Hongfu Sun, Yang Gao

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出SSDM-MRI方法，通过单步扩散模型和迭代选择性蒸馏算法实现高效MRI重建，具有快速重建时间和优异的图像质量。

Journal ref 2026

2506.17139 2026-01-15 cs.LG cs.AI physics.chem-ph physics.comp-ph stat.ML 78%

Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models

一致采样与模拟：基于能量的扩散模型的分子动力学

Michael Plainer, Hao Wu, Leon Klein, Stephan Günnemann, Frank Noé

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出了一种基于能量的扩散模型，通过福克-普朗克方程导出的正则化项来提高分子动力学模拟中的一致性和采样效率。

Comments Accepted at Conference on Neural Information Processing Systems (NeurIPS 2025)

2505.21400 2026-01-09 cs.LG cs.IT math.IT math.ST stat.ML stat.TH 78%

Breaking AR's Sampling Bottleneck: Provable Acceleration via Diffusion Language Models

突破生成模型的采样瓶颈：通过扩散语言模型实现可证明的加速

Gen Li, Changxiao Cai

机构 * Department of Statistics and Data Science, Chinese University of Hong Kong, Hong Kong（统计与数据科学系，香港中文大学）； Department of Industrial and Operations Engineering, University of Michigan, Ann Arbor, USA（工业与运营管理系，密歇根大学）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文从信息论角度为扩散语言模型提供收敛保证，证明采样误差随迭代次数减少而降低，从而突破自回归模型所需的L步瓶颈，为生成高质量样本提供理论支持。

Comments This is the full version of a paper published at NeurIPS 2025

2601.02236 2026-01-06 cs.CL 78%

CD4LM: Consistency Distillation and aDaptive Decoding for Diffusion Language Models

CD4LM：一致性蒸馏与自适应解码用于扩散语言模型

Yihao Liang, Ze Wang, Hao Chen, Ximeng Sun, Jialian Wu, Xiaodong Yu, Jiang Liu, Emad Barsoum, Zicheng Liu, Niraj K. Jha

机构 * Princeton University（普林斯顿大学）； Advanced Micro Devices, Inc（先进微器件公司）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结 CD4LM通过一致性蒸馏和自适应解码提升扩散语言模型的并行解码效率与生成质量。

Comments 33 pages, 7 figures

2512.12889 2025-12-16 cs.LG 78%

Distillation of Discrete Diffusion by Exact Conditional Distribution Matching

通过精确条件分布匹配实现离散扩散模型的蒸馏

Yansong Gao, Yu Sun

机构 * Google（谷歌）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结本文提出了一种基于条件分布匹配的蒸馏方法，用于加速离散扩散模型的推断过程，通过匹配教师和学生模型的条件分布来提高效率。

Comments [work in progress]

2405.15986 2025-12-01 cs.LG cs.DC cs.NA math.NA stat.ML 78%

Accelerating Diffusion Models with Parallel Sampling: Inference at Sub-Linear Time Complexity

通过并行采样加速扩散模型：在亚线性时间复杂度下进行推理

Haoxuan Chen, Yinuo Ren, Lexing Ying, Grant M. Rotskoff

机构 * ICME Stanford University（ICME 斯坦福大学）； Department of Mathematics and ICME Stanford University（数学系和ICME 斯坦福大学）； Department of Chemistry and ICME Stanford University（化学系和ICME 斯坦福大学）

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结通过并行采样技术，提出一种在亚线性时间复杂度下高效推理的扩散模型方法，实现了高维数据的快速高效采样。

Comments Accepted at NeurIPS 2024 as a Spotlight Poster (https://openreview.net/forum?id=F9NDzHQtOl)

2511.20705 2025-11-27 cs.LG cs.AI stat.ML 78%

Solving Diffusion Inverse Problems with Restart Posterior Sampling

利用重启后验抽样解决扩散逆问题

Bilal Ahmed, Joseph G. Makin

专题命中效率与蒸馏：diffusion(title,abstract)

AI总结 RePS通过重启后验抽样高效解决线性和非线性逆问题，避免反向传播，提升收敛速度和重建质量。

2511.14680 2025-11-19 eess.IV 78%

NERD: Network-Regularized Diffusion Sampling For 3D Computed Tomography

Shijun Liang, Ismail Alkhouri, Qing Qu, Rongrong Wang, Saiprasad Ravishankar

专题命中效率与蒸馏：diffusion(title,abstract)

Journal ref CAMSAP2025

2505.21005 2025-11-10 cs.LG stat.ML 78%

Efficient and Unbiased Sampling from Boltzmann Distributions via Variance-Tuned Diffusion Models

Fengzhe Zhang, Laurence I. Midgley, José Miguel Hernández-Lobato

机构 * University of Cambridge（剑桥大学）； Ångström AI（安格斯特罗姆人工智能）

专题命中效率与蒸馏：diffusion(title,abstract)

2505.19552 2025-11-07 cs.LG 78%

On scalable and efficient training of diffusion samplers

Minkyu Kim, Kiyoung Seong, Dongyeop Woo, Sungsoo Ahn, Minsu Kim

机构 * Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院）

专题命中效率与蒸馏：diffusion(title,abstract)

2505.13358 2025-10-24 cs.LG cs.AI 78%

One-Step Offline Distillation of Diffusion-based Models via Koopman Modeling

Nimrod Berman, Ilan Naiman, Moshe Eliasof, Hedi Zisling, Omri Azencot

机构 * Ben-Gurion University of the Negev（贝加尔-贡利亚大学）； University of Cambridge（剑桥大学）

专题命中效率与蒸馏：diffusion(title,abstract)

2212.11833 2025-10-23 econ.EM math.ST q-fin.RM stat.TH 78%

Efficient Sampling for Realized Variance Estimation in Time-Changed Diffusion Models

Timo Dimitriadis, Roxana Halbleib, Jeannine Polivka, Jasper Rennspies, Sina Streicher, Axel Friedrich Wolter

专题命中效率与蒸馏：diffusion(title,abstract)

2510.00627 2025-10-02 cs.AI 78%

Collaborative-Distilled Diffusion Models (CDDM) for Accelerated and Lightweight Trajectory Prediction

Bingzhang Wang, Kehua Chen, Yinhai Wang

机构 * Department of Civil and Environmental Engineering, University of Washington（华盛顿大学土木与环境工程系）

专题命中效率与蒸馏：diffusion(title,abstract)

2509.21470 2025-09-29 cs.LG cs.AI 78%

Score-based Idempotent Distillation of Diffusion Models

Shehtab Zaman, Chengyan Liu, Kenneth Chiu

专题命中效率与蒸馏：diffusion(title,abstract)

2401.02080 2025-09-26 cs.LG stat.CO stat.ML 78%

Energy based diffusion generator for efficient sampling of Boltzmann distributions

Yan Wang, Ling Guo, Hao Wu, Tao Zhou

机构 * School of Mathematical Sciences, Tongji University（同济大学数学科学学院）； Department of Mathematics, Shanghai Normal University（上海师范大学数学系）； Institute of Computational Mathematics and Scientific/Engineering Computing, AMSS, Chinese Academy of Sciences（中国科学院数学与系统科学研究院）

专题命中效率与蒸馏：diffusion(title,abstract)

2509.15865 2025-09-22 cs.LG 78%

SAGE: Semantic-Aware Shared Sampling for Efficient Diffusion

Haoran Zhao, Tong Bai, Lei Huang, Xiaoyu Liang

专题命中效率与蒸馏：diffusion(title,abstract)

Comments 5 pages, 4 figures

2508.20646 2025-08-29 cs.LG 78%

VarDiU: A Variational Diffusive Upper Bound for One-Step Diffusion Distillation

Leyang Wang, Mingtian Zhang, Zijing Ou, David Barber

机构 * University College London（伦敦大学学院）； Imperial College London（伦敦帝国学院）

专题命中效率与蒸馏：diffusion(title,abstract)

Comments Leyang Wang and Mingtian Zhang contributed equally to this work

2508.17868 2025-08-26 cs.SD cs.AI cs.LG eess.AS stat.ML 78%

FasterVoiceGrad: Faster One-step Diffusion-Based Voice Conversion with Adversarial Diffusion Conversion Distillation

Takuhiro Kaneko, Hirokazu Kameoka, Kou Tanaka, Yuto Kondo

机构 * NTT, Inc.（日本NTT公司）

专题命中效率与蒸馏：diffusion(title,abstract)

Comments Accepted to Interspeech 2025. Project page: https://www.kecl.ntt.co.jp/people/kaneko.takuhiro/projects/fastervoicegrad/

2508.16939 2025-08-26 cs.LG math.PR stat.ML 78%

Sig-DEG for Distillation: Making Diffusion Models Faster and Lighter

Lei Jiang, Wen Ge, Niels Cariou-Kotlarek, Mingxuan Yi, Po-Yu Chen, Lingyi Yang, Francois Buet-Golfouse, Gaurav Mittal, Hao Ni

机构 * University College London（伦敦大学学院）； JPMorganChase（摩根大通）； University of Oxford（牛津大学）； AIML Global Markets, Barclays（巴克莱证券AIML全球市场）

专题命中效率与蒸馏：diffusion(title,abstract)

2506.01083 2025-08-25 stat.ML cs.LG cs.SY eess.SY 78%

Generative diffusion posterior sampling for informative likelihoods

Zheng Zhao

机构 * Division of Statistics and Machine Learning（统计与机器学习系）

专题命中效率与蒸馏：diffusion(title,abstract)

Comments Commemorative issue for celebrating Thomas Kailath's 90th birthday

Journal ref Communications in Information and Systems, 2025

2507.15260 2025-07-22 cs.LG 78%

CHORDS: Diffusion Sampling Accelerator with Multi-core Hierarchical ODE Solvers

Jiaqi Han, Haotian Ye, Puheng Li, Minkai Xu, James Zou, Stefano Ermon

机构 * Stanford University（斯坦福大学）

专题命中效率与蒸馏：diffusion(title,abstract)

Comments ICCV 2025

2408.01519 2025-07-22 physics.med-ph 78%

Multi-Material Decomposition Using Spectral Diffusion Posterior Sampling

Xiao Jiang, Grace J. Gang, J. Webster Stayman

专题命中效率与蒸馏：diffusion(title,abstract)

Journal ref IEEE Transactions on Biomedical Engineering, vol. 72, no. 8, pp. 2447-2461, Aug. 2025

2507.11839 2025-07-17 cs.LG q-bio.QM 78%

Protenix-Mini: Efficient Structure Predictor via Compact Architecture, Few-Step Diffusion and Switchable pLM

Chengyue Gong, Xinshi Chen, Yuxuan Zhang, Yuxuan Song, Hao Zhou, Wenzhi Xiao

机构 * Tsinghua University（清华大学）

专题命中效率与蒸馏：diffusion(title,abstract)

2309.05019 2025-06-26 cs.LG stat.ML 78%

SA-Solver: Stochastic Adams Solver for Fast Sampling of Diffusion Models

Shuchen Xue, Mingyang Yi, Weijian Luo, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, Zhi-Ming Ma

机构 * University of Chinese Academy of Sciences（中国科学院大学）； Huawei Noah’s Ark Lab（华为诺亚实验室）； Peking University（北京大学）； Academy of Mathematics and Systems Science, Chinese Academy of Sciences（中国科学院数学与系统科学研究院）

专题命中效率与蒸馏：diffusion(title,abstract)

Comments Accepted in NeurIPS 2023

1. 效率与蒸馏 2537 篇

Accelerating Diffusion Planners in Offline RL via Reward-Aware Consistency Trajectory Distillation

SPREAD: Sampling-based Pareto front Refinement via Efficient Adaptive Diffusion

Ultra Fast PDE Solving via Physics Guided Few-step Diffusion

Beyond GEMM-Centric NPUs: Enabling Efficient Diffusion LLM Sampling

Enhanced posterior sampling via diffusion models for efficient metasurfaces inverse design

Variance-Reduced Diffusion Sampling via Target Score Identity

Highly Undersampled MRI Reconstruction via a Single Posterior Sampling of Diffusion Models

Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models

Breaking AR's Sampling Bottleneck: Provable Acceleration via Diffusion Language Models

CD4LM: Consistency Distillation and aDaptive Decoding for Diffusion Language Models

Distillation of Discrete Diffusion by Exact Conditional Distribution Matching

Accelerating Diffusion Models with Parallel Sampling: Inference at Sub-Linear Time Complexity

Solving Diffusion Inverse Problems with Restart Posterior Sampling

NERD: Network-Regularized Diffusion Sampling For 3D Computed Tomography

Efficient and Unbiased Sampling from Boltzmann Distributions via Variance-Tuned Diffusion Models

On scalable and efficient training of diffusion samplers

One-Step Offline Distillation of Diffusion-based Models via Koopman Modeling

Efficient Sampling for Realized Variance Estimation in Time-Changed Diffusion Models

Collaborative-Distilled Diffusion Models (CDDM) for Accelerated and Lightweight Trajectory Prediction

Score-based Idempotent Distillation of Diffusion Models

Energy based diffusion generator for efficient sampling of Boltzmann distributions

SAGE: Semantic-Aware Shared Sampling for Efficient Diffusion

VarDiU: A Variational Diffusive Upper Bound for One-Step Diffusion Distillation

FasterVoiceGrad: Faster One-step Diffusion-Based Voice Conversion with Adversarial Diffusion Conversion Distillation

Sig-DEG for Distillation: Making Diffusion Models Faster and Lighter

Generative diffusion posterior sampling for informative likelihoods

CHORDS: Diffusion Sampling Accelerator with Multi-core Hierarchical ODE Solvers

Multi-Material Decomposition Using Spectral Diffusion Posterior Sampling

Protenix-Mini: Efficient Structure Predictor via Compact Architecture, Few-Step Diffusion and Switchable pLM

SA-Solver: Stochastic Adams Solver for Fast Sampling of Diffusion Models