BundleNet: Learning with Noisy Label via Sample Correlations

BundleNet: Learning with Noisy Label via Sample Correlations
复制标题

BundleNet:通过样本相关性使用噪声标签进行学习

DOI:
10.1109/access.2017.2782844
复制
发表时间:
2018
期刊:
影响因子:
3.9
通讯作者:
Hanqing Lu
Hanqing Lu
中科院分区:
计算机科学3区
文献类型:
--
作者:
Chenghua Li;Chunjie Zhang;Kun Ding;Gang Li;Jian Cheng;Hanqing Lu

文献摘要

参考文献

被引文献

相似文献

序列模式很重要,因为它们可以被用来提高我们分类器的预测精度。在大数据量和深度学习的背景下,时序/视频帧和事件数据等时序数据在各种应用场景中变得越来越普遍。然而,用于训练现代机器学习模型的大数据集,如深度神经网络,经常受到标签噪声的影响。现有的噪声学习方法主要集中在构建一个额外的网络来去除噪声或寻找一个稳健的损失函数。很少有研究通过利用样本相关性来解决这个问题。在本文中,我们提出了BundleNet,一种用于深层神经网络处理标签噪声的顺序结构框架(称为BundleNet,称为BundleNet,见图1)。包模块通过逐类构建样本束来自然地考虑样本相关性,并将它们视为独立的输入。此外,我们还证明了束-模执行了一种形式的正则化,这类似于训练过程中的丢弃正则化。正则化效应使BundleNet对标签噪声具有较强的鲁棒性。在公开数据集上的大量实验证明了该方法的有效性和良好的应用前景。
Sequential patterns are important, because they can be exploited to improve the prediction accuracy of our classifiers. Sequential data, such as time series/video frames, and event data are becoming more and more ubiquitous in a wide spectrum of application scenarios especially in the background of large data and deep learning. However, large data sets used in training modern machine-learning models, such as deep neural networks, are often affected by label noise. Existing noisy learning approaches mainly focus on building an additional network to clean the noise or find a robust loss function. Few works tackle this problem by exploiting sample correlations. In this paper, we propose BundleNet, a framework of sequential structure (named bundle-module, see Fig. 1) for deep neural networks to handle the label noise. The bundle module naturally takes into account sample correlations by constructing bundles of samples class-by-class, and treats them as independent inputs. Moreover, we prove that the bundle-module performs a form of regularization, which is similar to dropout as regularization during training. The regularization effect endows the BundleNet with strong robustness to the label noise. Extensive experiments on public data sets prove that the proposed approach is effective and promising.
DOI: 10.1016/j.neunet.2017.01.003
发表时间: 2017-04
期刊: Neural networks : the official journal of the International Neural Network Society
影响因子: --
作者:
M. Pavel;Hannes Schulz;Sven Behnke
通讯作者: M. Pavel;Hannes Schulz;Sven Behnke
基于上下文样本分类器的图像表示进行分类
DOI: 10.1109/tcsvt.2016.2527380
发表时间: 2017-08
期刊: IEEE Transactions on Circuits System for Video Technology
影响因子: --
作者:
Chunjie Zhang;Qingming Huang;Qi Tian
通讯作者: Qi Tian
通过使用通用和特定类码本的低秩稀疏编码进行细粒度图像分类
DOI: 10.1109/tnnls.2016.2545112
发表时间: 2017-07
影响因子: 10.4
作者:
Chunjie Zhang;Chao Liang;Liang Li;Jing Liu;Qingming Huang;Qi Tian
通讯作者: Qi Tian
用于多标签学习的自定进度正则化框架
DOI: 10.1109/tnnls.2017.2697767
发表时间: 2018
影响因子: 10.4
作者:
Changsheng Li;Fan Wei;Junchi Yan;Xiaoyu Zhang;Qingshan Liu;Hongyuan Zha
通讯作者: Hongyuan Zha
DOI: 10.1016/j.ins.2016.06.029
发表时间: 2016-11
期刊: Inf. Sci.
影响因子: --
作者:
Chunjie Zhang;Zhe Xue;Xiaobin Zhu;Huanian Wang;Qingming Huang;Q. Tian
通讯作者: Chunjie Zhang;Zhe Xue;Xiaobin Zhu;Huanian Wang;Qingming Huang;Q. Tian