The expression is ambiguous, which is not convenient for readers to understand, and in today's view, the conclusion of the paper is of little significance, so it is no longer open
Data collection and annotation are time-consuming in machine learning, expecially for large scale problem. A common approach for this problem is to transfer knowledge from a related labeled domain to a target one. There are two popular ways to achieve this goal: adversarial learning and self training. In this article, we first analyze the training unstablity problem and the mistaken confusion issue in adversarial learning process. Then, inspired by domain confusion and self-ensembling methods, we propose a combined model to learn feature and class jointly invariant representation, namely Domain Confusion with Self Ensembling (DCSE). The experiments verified that our proposed approach can offer better performance than empirical art in a variety of unsupervised domain adaptation benchmarks.
CFEA: Collaborative Feature Ensembling Adaptation for Domain Adaptation in Unsupervised Optic Disc and Cup Segmentation
University of Florida(佛罗里达大学);UNC Charlotte(北卡罗来纳大学夏洛特分校);Xi’an Jiaotong University(西安交通大学);Sensetime Research(商汤科技研究院)
最近,深度神经网络在注释良好的数据集中表现出与董事会认证的眼科医生相当甚至更好的性能。然而,视网膜成像设备的多样性带来了一个重大挑战:域转移,这导致在将深度学习模型应用于新的测试域时性能下降。在本文中,我们提出了一种新的无监督域自适应框架,称为协作特征集成自适应(CFEA),以有效地克服这一挑战。我们提出的CFEA是一个交互式的范例,通过对抗学习和集成权重呈现出一种精致的协作适应。特别是,我们同时实现了域不变性,并保持了历史预测的指数移动平均值,这通过在训练期间集成权重来实现对未标记数据的更好预测。在不注释来自目标域的任何样本的情况下,编码器和解码器层中的多个对抗性损失引导域不变特征的提取,以混淆域分类器,同时有利于平滑权重的集成。综合实验结果表明,我们的CFEA模型可以克服性能下降,并优于国家的最先进的方法在分割视网膜视盘和杯眼底图像。\textit{Code is available at \url{https://github.com/cswin/AWC}}。
Do Ensembling and Meta-Learning Improve Outlier Detection in Randomized Controlled Trials?
Centre for Data Science and Digital Health, Hamilton Health Sciences(汉密尔顿健康科学中心数据科学与数字健康中心);Department of Statistical Sciences, University of Toronto(多伦多大学统计科学系);Institute for Health Policy, Management and Evaluation, University of Toronto(多伦多大学卫生政策、管理与评估研究所);Department of Medicine, McMaster University(麦克马斯特大学医学系);Population Health Research Institute(人口健康研究所)
虽然取得了显著的进展,但在没有任何标记数据的情况下,很难诱导出有监督的分类器。无监督的领域自适应能够通过将知识从标记的源域转移到未标记的目标域来克服这一挑战。可转移性和可区分性是表征特征表示的优越性的两个关键标准,以实现成功的领域自适应。在本文中,一种新的方法称为\textit{learning TransFerable and Discriminative Features for unsupervised domain adaptation}(TFDF)提出了这两个目标同时优化。一方面,进行分布对齐以减少域差异并学习更多的可传递表示。在再生核希尔伯特空间(RKHS)中,我们采用了一种新的统计量\textit{Maximum Mean and Covariance Discrepancy}(MMCD),它不仅能捕获一阶统计信息,还能捕获二阶统计信息,而不是采用仅捕获一阶统计信息的\textit{Maximum Mean Discrepancy}(MMD)来度量分布差异.另一方面,我们建议通过流形正则化探索局部判别信息,并通过最小化所提出的\textit{class confusion}目标来学习更多的判别特征,分别探索全局判别信息。我们将这两个目标集成到\textit{Structural Risk Minimization}(RSM)框架中,并学习域不变分类器。在5个真实数据集…
A Label Proportions Estimation Technique for Adversarial Domain Adaptation in Text Classification
Signal Analysis and Interpretation Lab(信号分析与解释实验室);University of Southern California(南加州大学);Department of Psychiatry and Behavioral Sciences(精神病学与行为科学系);University of Washington(华盛顿大学);Department Educational Psychology(教育心理学系);University of Utah(犹他大学)
SELF: Learning to Filter Noisy Labels with Self-Ensembling
Karlsruhe Institute of Technology(卡尔斯鲁厄理工学院);University of Freiburg(弗莱堡大学);Bosch GmbH(博世有限公司);Bosch Research, Bosch GmbH(博世有限公司博世研究院);Bosch Center for AI, Bosch GmbH(博世有限公司博世人工智能中心)