arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.11778cs.LGcs.AI
收藏

Towards Theoretical Understandings of Self-Consuming Generative Models(自消耗生成模型的理论理解研究)

Towards Theoretical Understandings of Self-Consuming Generative Models

University of Science and Technology of China(中国科学技术大学);The University of Sydney(悉尼大学);College of Control Science and Engineering, China University of Petroleum (East China)(中国石油大学(华东)控制科学与工程学院);Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, China(合肥综合性国家科学中心人工智能研究院);Nanyang Technological University(南洋理工大学)

Shi Fu, Sen Zhang, Yingjie Wang, Xinmei Tian, Dacheng Tao

AI总结:

该研究构建理论框架分析自消耗循环训练对生成模型学习分布的影响,推导了扩散模型的TV距离界,揭示合成数据量引发的相变现象,并给出核密度估计的相关结果。

摘要:

本文针对生成模型在自消耗循环中训练这一新兴挑战展开研究——在该循环中,连续多代模型会递归地用前代生成的真实数据与合成数据的混合集进行训练。我们构建了一套理论框架,可严格评估该训练流程对后续模型所学习的数据分布的影响,覆盖参数模型与非参数模型两类。具体而言,针对采用单隐层神经网络得分函数的diffusion models(扩散模型),我们推导了多种混合训练场景下,后续模型生成的合成数据分布与原始真实数据分布之间的全变分(TV)距离的界。分析表明,当混合训练数据集规模足够大、或真实数据占比足够高时,该距离可得到有效控制。有意思的是,我们进一步揭示了合成数据量扩张引发的相变现象:理论证明TV距离会先上升,在超过某个阈值点后转而下降。最后,我们给出了核密度估计的相关结果,提供了混合数据训练对误差传播影响等精细化洞见。

英文摘要:

This paper tackles the emerging challenge of training generative models within a self-consuming loop, wherein successive generations of models are recursively trained on mixtures of real and synthetic data from previous generations. We construct a theoretical framework to rigorously evaluate how this training procedure impacts the data distributions learned by future models, including parametric and non-parametric models. Specifically, we derive bounds on the total variation (TV) distance between the synthetic data distributions produced by future models and the original real data distribution under various mixed training scenarios for diffusion models with a one-hidden-layer neural network score function. Our analysis demonstrates that this distance can be effectively controlled under the condition that mixed training dataset sizes or proportions of real data are large enough. Interestingly, we further unveil a phase transition induced by expanding synthetic data amounts, proving theoretically that while the TV distance exhibits an initial ascent, it declines beyond a threshold point. Finally, we present results for kernel density estimation, delivering nuanced insights such as the impact of mixed data training on error propagation.

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。

↑