Dimensionality-Driven Learning with Noisy Labels

Autor:	Ma, Xingjun, Wang, Yisen, Houle, Michael E., Zhou, Shuo, Erfani, Sarah M., Xia, Shu-Tao, Wijewickrema, Sudanthi, Bailey, James
Rok vydání:	2018
Předmět:	Computer Science - Computer Vision and Pattern Recognition Computer Science - Machine Learning Statistics - Machine Learning
Druh dokumentu:	Working Paper
Popis:	Datasets with significant proportions of noisy (incorrect) class labels present challenges for training accurate Deep Neural Networks (DNNs). We propose a new perspective for understanding DNN generalization for such datasets, by investigating the dimensionality of the deep representation subspace of training samples. We show that from a dimensionality perspective, DNNs exhibit quite distinctive learning styles when trained with clean labels versus when trained with a proportion of noisy labels. Based on this finding, we develop a new dimensionality-driven learning strategy, which monitors the dimensionality of subspaces during training and adapts the loss function accordingly. We empirically demonstrate that our approach is highly tolerant to significant proportions of noisy labels, and can effectively learn low-dimensional local subspaces that capture the data distribution. Comment: In Proceedings of the International Conference on Machine Learning (ICML), 2018
Databáze:	arXiv
Externí odkaz:	http://arxiv.org/abs/1806.02612 Zobrazit plný text záznamu View this record from Arxiv