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 2302.08883stat.MLcs.AIcs.LGstat.ME
收藏

近似贝叶斯最优的伪标签选择

Approximately Bayes-Optimal Pseudo Label Selection

Ludwig-Maximilians-Universität (LMU) Munich(慕尼黑路德维希-马克西米利安大学(慕尼黑大学));Munich Center for Machine Learning (MCML)(慕尼黑机器学习中心);Fraunhofer Institute of Integrated Circuits (IIS)(弗劳恩霍夫集成电路研究所);Helmholtz Zentrum Neuherberg(亥姆霍兹纽赫贝格中心)

Julian Rodemann, Jann Goschenhofer, Emilio Dorigatti, Thomas Nagler, Thomas Augustin

AI总结:

本文提出贝叶斯伪标签选择框架BPLS,通过解析近似伪样本后验预测来缓解自训练中的确认偏差,并在高维数据上优于传统方法。

摘要:

通过自训练进行的半监督学习严重依赖伪标签选择(PLS)。这种选择通常依赖于在已标注数据上拟合的初始模型。因此,早期的过拟合可能会通过选择具有过度自信但错误预测的实例而传播到最终模型中,这通常被称为确认偏差。本文引入了 BPLS,一个旨在缓解此问题的 PLS 贝叶斯框架。其核心在于选择实例进行标注的标准:伪样本后验预测的解析近似。我们通过证明伪样本后验预测的贝叶斯最优性来推导该选择标准。我们进一步通过解析近似该标准来克服计算障碍。其与边际似然的关系使我们能够提出一种基于 Laplace 方法和 Gaussian 积分的近似。我们在模拟和真实数据上,针对参数化广义线性和非参数化广义加性模型对 BPLS 进行了实证评估。在面对容易过拟合的高维数据时,BPLS 优于传统的 PLS 方法。

英文摘要:

Semi-supervised learning by self-training heavily relies on pseudo-label selection (PLS). The selection often depends on the initial model fit on labeled data. Early overfitting might thus be propagated to the final model by selecting instances with overconfident but erroneous predictions, often referred to as confirmation bias. This paper introduces BPLS, a Bayesian framework for PLS that aims to mitigate this issue. At its core lies a criterion for selecting instances to label: an analytical approximation of the posterior predictive of pseudo-samples. We derive this selection criterion by proving Bayes optimality of the posterior predictive of pseudo-samples. We further overcome computational hurdles by approximating the criterion analytically. Its relation to the marginal likelihood allows us to come up with an approximation based on Laplace's method and the Gaussian integral. We empirically assess BPLS for parametric generalized linear and non-parametric generalized additive models on simulated and real-world data. When faced with high-dimensional data prone to overfitting, BPLS outperforms traditional PLS methods.

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。

↑