Timestep

Follow this topic Ask about this topic

Momentum

1 paper in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.

Jul 13Week of Sep 28

Latest papers 27

All topics

Principal-timestep Restricted Init via Sparse Matrix-decomposition in Flow-matching
Sep 14, 2026Jiayang Gu, Zheng Fang, Lichaun Xiang +3Timestep Autoregressive Language Models
University of Warwick · Bytedance
Time Without Timesteps: Simulating Coupled Dynamical Systems via Self-Consistency
Sep 3, 2026Liyu Zerihun, Mark Shinyoung LeeDynamical Systems Slow Relaxation Modes
Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning
Sep 1, 2026Mariia Drozdova, Aidan Sirbu, Pietro Miotti +4Functional Code Solvers Denoising Diffusion Probabilistic Model
1Google, Paradigms of Intelligence Team · School of Computer Science, McGill University · 4Mila - Quebec AI Institute +3
Latent-Centroid Steering: Single-Pass Classifier-Free Guidance for Command-Aligned Autonomous Driving
Jul 31, 2026Meibo Hu, Jiamian Wang, Pichao Wang +1Autonomous Driving Steering
Rochester Institute of Technology · NVIDIA
TARS: Timestep-Aware Data Scaling for 3D-Free Video Re-Shooting
Jul 30, 2026Jiwen Liu, Shujuan Li, Xiaohan Li +5Scene Reconstruction Camera
1Kling Team · 2Tsinghua University · 3Peking University
TPD: Temporal Prior Decoupling for Text-to-Video Diffusion Models
Jul 29, 2026Taewon Kang, Matthias ZwickerText-To-Video Diffusion Models Diffusion Models
University of Maryland at College Park, United States
Reflex: Real-Time VLA Control through Streaming Inference
Jul 16, 2026Yuanchun Guo, Bingyan LiuFlow-Matching Vision-Language-Action Streaming Inference
School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China.
Selective Timestep Weighting and Advantage-Based Replay for Sample-Efficient Diffusion RLHF
Jul 8, 2026Eric Zhu, Abhinav Shrivastava, Soumik MukhopadhyayDiffusion-Based Reinforcement Learning Methods Reinforcement Learning From Human Feedback
Carnegie Mellon University · University of Maryland, College Park
Adaptive Reparametrized Time for Score-Based Diffusion Sampling
Jul 2, 2026Yilie Huang, Wenpin Tang, Xun Yu ZhouDiffusion Sampling Timestep
Department of Applied Mathematics, The Hong Kong Polytechnic University, Hong Kong. · Department of Industrial Engineering and Operations Research, Columbia University, New York, NY 10027, USA. · Department of Industrial Engineering and Operations Research and Data Science Institute, Columbia University, New York, NY 10027, USA.
Subliminal Clocks: Latent Time Modelling in Diffusion Language Models
Jul 2, 2026Maximo Rulli, Thomas Fontanari, Simone Petruzzi +9Diffusion Language Models Timestep
Sapienza University of Rome · EPFL · Independent researcher
MUSE: Unlocking Timestep as Native Task Steering for One-Step Dense Prediction
Jun 29, 2026Shuo Zhou, Zhaoxin Li, Xiujuan ChaiDense Prediction Multi--Task Learning
Agricultural Information Institute, Chinese Academy of Agricultural Sciences, Beijing, China. · Key Laboratory of Agricultural Big Data, Ministry of Agriculture and Rural Affairs, Beijing, China.
Timesteps of Mamba Align with Human Reading Times
Jun 29, 2026Yuji Yamamoto, Shinnosuke Isono, Yoshinobu Kawahara +1Mamba Language Modeling
SOKENDAI · NINJAL · The University of Osaka +2
Accelerated sampling using SamAdams variable timesteps and position-adaptive Langevin dynamics
Jun 25, 2026Benedict Leimkuhler, Peter A. WhalleyLangevin Dynamics Accelerated Sampling
School of Mathematics, University of Edinburgh, Edinburgh, UK. · Seminar for Statistics, Department of Mathematics, ETH Zürich, Zürich, Switzerland.
A Causal DAG Prior for Synthetic Time-Series Classification Datasets
Jun 19, 2026Franco Martino O'Rourke, Ana Trisovic, Dimitris BertsimasTabular Prior-Data Fitted Network Time-Series Classification
Massachusetts Institute of Technology, Cambridge, MA, USA.
Beyond Tokenization: Direct Timestep Embedding and Contrastive Alignment for Time-Series Question Answering
Jun 17, 2026Yafeng Wu, Huu Hiep Nguyen, Thin Nguyen +1Timestep Embeddings Time Series
Deakin University
When Dynamics Models Read the Wrong Time Steps: Label-Free Event Credit Re-Anchoring for Robust Global Readouts
Jun 16, 2026Yifan WangInference Dynamics Readout
Department of Mechanical Engineering, McGill University Montréal, QC H3A 2T7, Canada
Unlocking Diffusion Hierarchies: Adaptive Timestep Selection for Zero-Shot Segmentation
Jun 14, 2026Ramin Nakhli, Mahesh Ramachandran, Luca BallanFew-Shot Segmentation Diffusion Models
Google
Timestep Rescheduling in Diffusion Inversion
Jun 13, 2026Shangquan Sun, Ting Gong, Zhirui Liu +5Diffusion Inversion Timestep
School of CyberScience and Technology, Sun Yat-sen University, Shenzhen Campus, Shenzhen 518107, China · Tsinghua University · CUHK +2
Rethinking One-Step Image Editing through ChordEdit: Reproduction, Simplification, and New Insights
Jun 12, 2026Minghan Li, Jeremy Moebel, Mengyu WangImage Editing Timestep
Harvard AI and Robotics Lab
Tracing the Oracle: Improving Diffusion Timestep Scheduling for 3D CT Reconstruction
Jun 4, 2026Yujia Wu, Zhaoqiang LiuTimestep Cone-Beam Computed Tomography
The authors are with the School of Computer Science and Engineering, University of Electronic Science and Technology of China.
Not All Timesteps Matter Equally: Selective Alignment Knowledge Distillation for Spiking Neural Networks
May 14, 2026Kai Sun, Peibo Duan, Yongsheng Huang +4Spiking Neural Networks Knowledge Distillation
Faculty of Information Technology, Monash University, Australia · School of Software, Northeastern University, China · Department of Medicine, National University of Singapore, Singapore
Denoising, Fast and Slow: Difficulty-Aware Adaptive Sampling for Image Generation
Apr 21, 2026Johannes Schusterbauer, Ming Gui, Yusong Li +3Image Generation Timestep
Elucidating the SNR-t Bias of Diffusion Probabilistic Models
Apr 17, 2026Meng Yu, Lei Sun, Jianhao Zeng +2Denoising Diffusion Probabilistic Model Diffusion Models
Lanzhou University · AMAP Alibaba Group
SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching
Feb 27, 2026Yasaman Haghighi, Alexandre AlahiDiffusion Models Cache
Ecole Polytechnique Fédérale de Lausanne (EPFL)
Formalizing the Sampling Design Space of Diffusion-Based Generative Models via Adaptive Solvers and Wasserstein-Bounded Timesteps
Feb 13, 2026Sangwoo Jo, Sungjoon ChoiDiffusion Sampling Timestep
Department of Artificial Intelligence, Korea University, Seoul, South Korea.
ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Schedule
Jan 26, 2026Yilie Huang, Wenpin Tang, Xunyu ZhouTimestep Diffusion Sampling
Department of Industrial Engineering and Operations Research, Columbia University, New York, NY 10027, USA. · Department of Industrial Engineering and Operations Research & Data Science Institute, Columbia University, New York, NY 10027, USA.
Adaptive Time-step Training for Enhancing Spike-Based Neural Radiance Fields
Jul 30, 2025Ranxi Lin, Canming Yao, Jiayi Li +3Neural Radiance Fields Timestep
School of Information Science and Technology, ShanghaiTech University, Shanghai, China · Guangdong Institute of Intelligence Science and Technology, Zhuhai, China

Sep 14, 2026, cs.LGJ/K move · Enter open · S save

Principal-timestep Restricted Init via Sparse Matrix-decomposition in Flow-matching

Jiayang Gu, Zheng Fang, Lichaun Xiang, Fanghui Liu+2

University of Warwick · Bytedance

Timestep Autoregressive Language Models

Open paper AskPDF arXiv

Flow-matching diffusion models have recently emerged as a strong paradigm for high-fidelity visual generation. However, their prohibitively high fine-tuning cost limits scalability to downstream tasks. While Low-Rank Adaptation (LoRA) combined with spectral initialization has demonstrated accelerated convergence and improved performance in autoregressive language models by better aligning gradient directions, we find that it fails to deliver similar gains in diffusion fine-tuning, often yielding marginal or even negative improvements over vanilla LoRA.We attribute this discrepancy to a fundamental mismatch between LoRA's low-rank parameterization and the intrinsically high-rank gradients induced by the flow-matching objective. In particular, stochastic timestep sampling introduces directionally heterogeneous gradient signals across training steps, leading to misaligned updates under low-rank constraints.To address this issue, we propose Prism-LoRA,a Principal-timestep Restricted Init via Sparse Matrix-decomposition framework that improves gradient alignment during fine-tuning. Our method consists of two key components: (i) principal timestep selection, which restricts initialization gradients to a subset of dominant timesteps to suppress effective gradient rank, and (ii) principal channel filtering, which removes task-irrelevant channels, enabling the one-step spectral initialization gradient to better align with the long-horizon optimization trajectory. Extensive experiments demonstrate that our method consistently improves both convergence speed and final performance across multiple diffusion fine-tuning benchmarks, including subject-driven generation, controllable generation, and deblurring, achieving not only performance improvement but also earlier stages of convergence over baseline LoRA and other spectral-init methods.

Momentum

Related topics

Latest papers 27