Synergistic Network Learning and Label Correction for Noise-robust Image Classification
This work addresses label noise in image classification, a common issue in real-world datasets, but it is incremental as it builds on existing ideas like small loss selection and noise correction.
The paper tackles the problem of label noise in large training datasets, which degrades deep neural network performance, by proposing a robust label correction framework that iteratively learns network parameters and reassigns ground truth labels, achieving superior results on synthetic and real-world datasets like CIFAR-10, CIFAR-100, and Clothing1M.
Large training datasets almost always contain examples with inaccurate or incorrect labels. Deep Neural Networks (DNNs) tend to overfit training label noise, resulting in poorer model performance in practice. To address this problem, we propose a robust label correction framework combining the ideas of small loss selection and noise correction, which learns network parameters and reassigns ground truth labels iteratively. Taking the expertise of DNNs to learn meaningful patterns before fitting noise, our framework first trains two networks over the current dataset with small loss selection. Based on the classification loss and agreement loss of two networks, we can measure the confidence of training data. More and more confident samples are selected for label correction during the learning process. We demonstrate our method on both synthetic and real-world datasets with different noise types and rates, including CIFAR-10, CIFAR-100 and Clothing1M, where our method outperforms the baseline approaches.