# Generalization in Generation: A closer look at Exposure Bias

arXiv: https://arxiv.org/abs/1910.00292 · HTML: https://mlsanity.com/paper/1910.00292/

## Key facts

- Cited by 123 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Oct 2019.

## Built on by (most cited first, 30 of 123)

- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Vector Quantized Diffusion Model for Text-to-Image Synthesis](https://mlsanity.com/paper/2111.14822/) (2021)
- [Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion](https://mlsanity.com/paper/2506.08009/) (2025)
- [Diffsound: Discrete Diffusion Model for Text-to-sound Generation](https://mlsanity.com/paper/2207.09983/) (2022)
- [Rolling Forcing: Autoregressive Long Video Diffusion in Real Time](https://mlsanity.com/paper/2509.25161/) (2025)
- [Factuality Challenges in the Era of Large Language Models](https://mlsanity.com/paper/2310.05189/) (2023)
- [ImageBART: Bidirectional Context with Multinomial Diffusion for Autoregressive Image Synthesis](https://mlsanity.com/paper/2108.08827/) (2021)
- [Why Exposure Bias Matters: An Imitation Learning Perspective of Error Accumulation in Language Generation](https://mlsanity.com/paper/2204.01171/) (2022)
- [Elucidating the Exposure Bias in Diffusion Models](https://mlsanity.com/paper/2308.15321/) (2023)
- [State-of-the-art generalisation research in NLP: A taxonomy and review](https://mlsanity.com/paper/2210.03050/) (2022)
- [Fast Unsupervised Brain Anomaly Detection and Segmentation with Diffusion Models](https://mlsanity.com/paper/2206.03461/) (2022)
- [On-Policy RL Meets Off-Policy Experts: Harmonizing Supervised Fine-Tuning and Reinforcement Learning via Dynamic Weighting](https://mlsanity.com/paper/2508.11408/) (2025)
- [Input Perturbation Reduces Exposure Bias in Diffusion Models](https://mlsanity.com/paper/2301.11706/) (2023)
- [Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps](https://mlsanity.com/paper/2305.15583/) (2023)
- [Magic-Me: Identity-Specific Video Customized Diffusion](https://mlsanity.com/paper/2402.09368/) (2024)
- [End-to-End Training for Autoregressive Video Diffusion via Self-Resampling](https://mlsanity.com/paper/2512.15702/) (2025)
- [G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model](https://mlsanity.com/paper/2208.09141/) (2022)
- [Tuning computer vision models with task rewards](https://mlsanity.com/paper/2302.08242/) (2023)
- [Exploiting the Signal-Leak Bias in Diffusion Models](https://mlsanity.com/paper/2309.15842/) (2023)
- [MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods](https://mlsanity.com/paper/2309.10966/) (2023)
- [Reward Gaming in Conditional Text Generation](https://mlsanity.com/paper/2211.08714/) (2022)
- [Taming Teacher Forcing for Masked Autoregressive Video Generation](https://mlsanity.com/paper/2501.12389/) (2025)
- [ShotStream: Streaming Multi-Shot Video Generation for Interactive Storytelling](https://mlsanity.com/paper/2603.25746/) (2026)
- [Causal-rCM: A Unified Teacher-Forcing and Self-Forcing Open Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation and Interactive World Models](https://mlsanity.com/paper/2606.25473/) (2026)
- [Coarse-to-Fine: a Hierarchical Diffusion Model for Molecule Generation in 3D](https://mlsanity.com/paper/2305.13266/) (2023)
- [Relating Neural Text Degeneration to Exposure Bias](https://mlsanity.com/paper/2109.08705/) (2021)
- [HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration](https://mlsanity.com/paper/2410.01723/) (2024)
- [Implicit Unlikelihood Training: Improving Neural Text Generation with Reinforcement Learning](https://mlsanity.com/paper/2101.04229/) (2021)
- [A Systematic Study of Knowledge Distillation for Natural Language Generation with Pseudo-Target Training](https://mlsanity.com/paper/2305.02031/) (2023)
- [GRIFFIN: Effective Token Alignment for Faster Speculative Decoding](https://mlsanity.com/paper/2502.11018/) (2025)

## Abstract (arXiv)

Exposure bias refers to the train-test discrepancy that seemingly arises when an autoregressive generative model uses only ground-truth contexts at training time but generated ones at test time. We separate the contributions of the model and the learning framework to clarify the debate on consequences and review proposed counter-measures. In this light, we argue that generalization is the underlying property to address and propose unconditional generation as its fundamental benchmark. Finally, we combine latent variable modeling with a recent formulation of exploration in reinforcement learning to obtain a rigorous handling of true and generated contexts. Results on language modeling and variational sentence auto-encoding confirm the model's generalization capability.