Exposure Bias

Follow this topic Ask about this topic

Latest papers

10 papers0 papers in the last four weeks, with none the four weeks before. 0.0% of all new papers.

Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matching
Jun 26, 2026Guanbo Huang, Jingjia Mao, Fanding Huang +9Flow Matching Image Generation
Tsinghua Shenzhen International Graduate School, Tsinghua University · Central Media Technology Institute, Huawei
Exposure Bias as Epistemic Underidentification in Recursive Forecasting
Jun 11, 2026Riku Green, Zahraa S. Abdallah, Telmo M Silva FilhoDistribution Shift Time Series Forecasting
University of Bristol, UK
IMUG-Bench: Benchmarking Unified Multimodal Models on Interleaved Understanding and Generation
Jun 8, 2026Lingyi Meng, Zecong Tang, Haoran Li +12Unified Multimodal Models Multimodal Generation
Zhejiang University · Institute of Automation, Chinese Academy of Sciences · The University of Hong Kong +1
The Bridge-Garden Dilemma in LLM Distillation: Why Mixing Hard and Soft Labels Works
May 25, 2026Guanghui Wang, Kaiwen Lv Kacuila, Zhiyong Yang +5LLM Compression Language Model Distillation
School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China · Alibaba Group, Hangzhou, China · State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China +2
Distribution Corrected Offline Data Distillation for Large Language Models
May 13, 2026Yumeng Zhang, Zhengbang Yang, Yevin Nikhel Goonatilake +1CoT Distillation Online Knowledge Distillation
George Mason University, Fairfax, VA, USA
COMO: Closed-Loop Optical Molecule Recognition with Minimum Risk Training
Apr 26, 2026Zhuoqi Lyu, Qing KeExposure Bias Optical Character Recognition
Department of Data Science, College of Computing, City University of Hong Kong, Hong Kong, China
RAE-AR: Taming Autoregressive Models with Representation Autoencoders
Apr 2, 2026Hu Yu, Hang Xu, Jie Huang +4Exposure Bias Autoregressive Image Generation
University of Science and Technology of China · JD Explore Academy · The University of Hong Kong
A Survey of On-Policy Distillation for Large Language Models
Apr 1, 2026Mingyang Song, Mao ZhengExposure Bias Language Model Distillation
Large Language Model Department, Tencent, China
Generate Then Correct: Single Shot Global Correction for Aspect Sentiment Quad Prediction
Mar 14, 2026Shidong He, Haoyu Wang, Wenjie LuoExposure Bias Aspect-Based Sentiment Analysis
School of Cyber Security and Computer, Hebei University, Baoding, China. · Faculty of Computer Science, Dalhousie University, Halifax, Canada.
Debiasing Sequential Recommendation with Time-aware Inverse Propensity Scoring
Mar 5, 2026Sirui Huang, Jing Long, Qian Li +2Exposure Bias Sequential Recommendation
Hong Kong Polytechinic University Hong Kong SAR, China · University of Technology Sydney Sydney, Australia · Curtin University Perth, Australia +1

Jun 26, 2026J/K move · Enter open · S save

Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matching

Guanbo Huang, Jingjia Mao, Fanding Huang, Fengkai Liu+8

Tsinghua Shenzhen International Graduate School, Tsinghua University · Central Media Technology Institute, Huawei

Flow Matching Image Generation Exposure Bias

Open paper AskPDF arXiv

Flow Matching (FM) has achieved remarkable generative performance, yet it suffers from exposure bias due to discrepancies between training and inference. Existing mitigation strategies typically rely on static constraints or external heuristics. In this work, we propose that exposure bias itself inherently contains dynamic signals that can guide its own rectification. To leverage this, we introduce DEFAR (DirEctional-Frequency Adaptive Rectification). This framework simulates the single-step inference process during training to identify exposure bias. It utilizes directional and frequency-adaptive feedback signals from the bias itself to enhance the model's bias tolerance. It consists of two key components: (1) Anti-Drift Rectification (ADR). ADR treats inference-time drift as a signal to learn the direction to steer deviated states back toward the target. ADR endows the model with intrinsic active self-rectification capabilities; (2) Frequency Compensation (FC). Empirically, we observe that accumulated bias often stems from a lack of low-frequency components in high-noise stages, and exposure bias carries the missing frequency. FC leverages the bias itself as a self-feedback weighting factor to reinforce the missing frequency components. Experiments on CIFAR-10, CelebA-64, and ImageNet-256/512 show that DEFAR outperforms prior baselines and further demonstrates favorable scalability, compatibility, and inference robustness.