Summary of Confronting Reward Overoptimization For Diffusion Models: a Perspective Of Inductive and Primacy Biases, by Ziyi Zhang and Sen Zhang and Yibing Zhan and Yong Luo and Yonggang Wen and Dacheng Tao
Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases
by Ziyi Zhang, Sen Zhang, Yibing Zhan, Yong Luo, Yonggang Wen, Dacheng Tao
First submitted to arxiv on: 13 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper addresses the issue of optimizing reward models for diffusion models to align with human preferences. While current methods optimize downstream rewards, concerns arise about compromising ground-truth performance. The authors identify a mismatch between existing methods and the temporal inductive bias of diffusion models, leading to overoptimization risks. They also discover that dormant neurons in the critic model serve as regularization against reward overoptimization. To mitigate this issue, they propose Temporal Diffusion Policy Optimization with critic active neuron Reset (TDPO-R), a policy gradient algorithm leveraging the temporal inductive bias and primacy bias. Experimental results demonstrate the superior effectiveness of their method in mitigating reward overoptimization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps bridge the gap between diffusion models and human preferences, making them more useful for practical generative tasks. It identifies problems with current methods that optimize rewards and suggests a new approach to fix these issues. The authors use special algorithms and techniques to make their method work better. |
Keywords
* Artificial intelligence * Diffusion * Optimization * Regularization
