Auxiliary Image Regularization for Deep CNNs with Noisy Labels
arXiv:1511.07069
Abstract
Precisely-labeled data sets with sufficient amount of samples are very important for training deep convolutional neural networks (CNNs). However, many of the available real-world data sets contain erroneously labeled samples and those errors substantially hinder the learning of very accurate CNN models. In this work, we consider the problem of training a deep CNN model for image classification with mislabeled training samples - an issue that is common in real image data sets with tags supplied by amateur users. To solve this problem, we propose an auxiliary image regularization technique, optimized by the stochastic Alternating Direction Method of Multipliers (ADMM) algorithm, that automatically exploits the mutual context information among training images and encourages the model to select reliable images to robustify the learning process. Comprehensive experiments on benchmark data sets clearly demonstrate our proposed regularized CNN model is resistant to label noise in training data.
Published as a conference paper at ICLR 2016
References in corpus (4)
Cited by in corpus (10)
- Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels
- SELF: Learning to Filter Noisy Labels with Self-Ensembling
- Uncertainty Based Detection and Relabeling of Noisy Image Labels
- Learning Deep Networks from Noisy Labels with Dropout Regularization
- Robust Learning Under Label Noise With Iterative Noise-Filtering
- On Universal Black-Box Domain Adaptation
- Learning from Web Data: the Benefit of Unsupervised Object Localization
- A Simple yet Effective Baseline for Robust Deep Learning with Noisy Labels
- Facial Emotion Recognition with Noisy Multi-task Annotations
- Limited Gradient Descent: Learning With Noisy Labels