arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2502.09963cs.CV
收藏

基于生成内容再生成:一种迈向自进化扩散模型的方法

Generating on Generated: An Approach Towards Self-Evolving Diffusion Models

The Hong Kong Polytechnic University(香港理工大学);CAIR, HKISI, CAS(中国科学院香港中文大学深圳先进集成技术研究所);UCAS(中国科学院大学);CASIA(中国科学院自动化研究所)

Xulu Zhang, Xiaoyong Wei, Jinlin Wu, Jiaxin Wu, Zhaoxiang Zhang, Zhen Lei, Qing Li

AI总结:

针对文本到图像扩散模型使用合成数据自训练时的崩溃问题,本文提出提示词过滤、偏好采样和分布加权三项策略,以提升感知对齐并抑制幻觉累积。

摘要:

递归自我改进(Recursive Self-Improvement,RSI)使智能系统能够自主完善自身能力。本文探索 RSI 在文本到图像扩散模型中的应用,以应对由合成数据导致的训练崩溃挑战。我们发现造成这种崩溃的两个关键因素:感知对齐缺失,以及生成幻觉的累积。为缓解这些问题,我们提出三种策略:(1)一套提示词构建与过滤流程,旨在促进感知对齐数据的生成;(2)一种偏好采样方法,用于识别人类偏好的样本并过滤生成幻觉;(3)一种基于分布的加权方案,用于惩罚带有幻觉错误的入选样本。大量实验验证了这些方法的有效性。

英文摘要:

Recursive Self-Improvement (RSI) enables intelligence systems to autonomously refine their capabilities. This paper explores the application of RSI in text-to-image diffusion models, addressing the challenge of training collapse caused by synthetic data. We identify two key factors contributing to this collapse: the lack of perceptual alignment and the accumulation of generative hallucinations. To mitigate these issues, we propose three strategies: (1) a prompt construction and filtering pipeline designed to facilitate the generation of perceptual aligned data, (2) a preference sampling method to identify human-preferred samples and filter out generative hallucinations, and (3) a distribution-based weighting scheme to penalize selected samples with hallucinatory errors. Our extensive experiments validate the effectiveness of these approaches.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑