arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2202.00796cs.LGcs.ITmath.IT
收藏

论无监督多源无源域自适应中的偏差与方差平衡

On Balancing Bias and Variance in Unsupervised Multi-Source-Free Domain Adaptation

Massachusetts Institute of Technology(麻省理工学院);University of Florida(佛罗里达大学)

Maohao Shen, Yuheng Bu, Gregory Wornell

AI总结:

本文从信息论角度分析多源无源域自适应的泛化误差极限,揭示偏差-方差权衡,并提出域聚合、选择性伪标注和联合特征对齐三种平衡策略及新算法,在多个数据集上取得最先进性能。

摘要:

由于隐私、存储及其他约束,机器学习中越来越需要不要求访问用于训练一组源模型的数据的无监督域自适应技术。现有的多源无源域自适应(MSFDA)方法通常利用源模型生成的伪标注数据来训练目标模型,这些方法侧重于改进伪标注技术或提出新的训练目标。相反,我们旨在分析 MSFDA 的基本极限。特别地,我们针对所得目标模型的泛化误差建立了一个信息论界,该界揭示了固有的偏差-方差权衡。随后,我们从三个视角提供了如何平衡这一权衡的见解,包括域聚合、选择性伪标注和联合特征对齐,从而设计出新算法。在多个数据集上的实验验证了我们的理论分析,并证明了所提算法的最先进性能,尤其是在一些最具挑战性的数据集上,包括 Office-Home 和 DomainNet。

英文摘要:

Due to privacy, storage, and other constraints, there is a growing need for unsupervised domain adaptation techniques in machine learning that do not require access to the data used to train a collection of source models. Existing methods for multi-source-free domain adaptation (MSFDA) typically train a target model using pseudo-labeled data produced by the source models, which focus on improving the pseudo-labeling techniques or proposing new training objectives. Instead, we aim to analyze the fundamental limits of MSFDA. In particular, we develop an information-theoretic bound on the generalization error of the resulting target model, which illustrates an inherent bias-variance trade-off. We then provide insights on how to balance this trade-off from three perspectives, including domain aggregation, selective pseudo-labeling, and joint feature alignment, which leads to the design of novel algorithms. Experiments on multiple datasets validate our theoretical analysis and demonstrate the state-of-art performance of the proposed algorithm, especially on some of the most challenging datasets, including Office-Home and DomainNet.

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。

↑