转移与边缘化：用模糊信息解释标签噪声
Transfer and Marginalize: Explaining Away Label Noise with Privileged Information
摘要：
监督学习数据集通常具有特权信息，其形式为在训练时可用但在测试时不可用的特征，例如提供标签的注释者的ID。我们认为，特权信息是有用的解释远离标签噪音，从而减少噪音标签的有害影响。我们开发了一个简单而有效的方法，监督学习与神经网络：它通过权重共享知识学习特权信息和近似边缘化特权信息在测试时。我们的方法，TRAM（TRansfer and Marginalize），具有最小的训练时间开销，并且具有与不使用特权信息相同的测试时间成本。TRAM在CIFAR-10 H、ImageNet和Civil Comments基准测试中表现强劲。
英文摘要：
Supervised learning datasets often have privileged information, in the form of features which are available at training time but are not available at test time e.g. the ID of the annotator that provided the label. We argue that privileged information is useful for explaining away label noise, thereby reducing the harmful impact of noisy labels. We develop a simple and efficient method for supervised learning with neural networks: it transfers via weight sharing the knowledge learned with privileged information and approximately marginalizes over privileged information at test time. Our method, TRAM (TRansfer and Marginalize), has minimal training time overhead and has the same test-time cost as not using privileged information. TRAM performs strongly on CIFAR-10H, ImageNet and Civil Comments benchmarks.
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。
