You are here

Unsupervised label noise modeling and loss correction

Publication Type: 
Refereed Conference Meeting Proceeding
Abstract: 
Despite being robust to small amounts of label noise, convolutional neural networks trained with stochastic gradient methods have been shown to easily fit random labels. When there are a mixture of correct and mislabelled targets, networks tend to fit the former before the latter. This suggests using a suitable two-component mixture model as an unsupervised generative model of sample loss values during training to allow online estimation of the probability that a sample is mislabelled. Specifically, we propose a beta mixture to estimate this probability and correct the loss by relying on the network prediction (the so-called bootstrapping loss). We further adapt mixup augmentation to drive our approach a step further. Experiments on CIFAR-10/100 and TinyImageNet demonstrate a robustness to label noise that substantially outperforms recent state-of-the-art.
Conference Name: 
36th International Conference on Machine Learning. International Conference on Machine Learning
Digital Object Identifer (DOI): 
10.23274
Publication Date: 
10/06/2019
Conference Location: 
United States of America
Research Group: 
Institution: 
Dublin City University (DCU)
Open access repository: 
Yes