Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks

TL;DR

This paper introduces a semi-supervised framework using an undirected graphical model to improve training of deep neural networks with noisy labels, enhancing label accuracy and robustness.

Contribution

It presents a novel probabilistic model that jointly infers clean labels and trains neural networks, effectively handling label noise in large datasets.

Findings

Reduces label noise in CIFAR-10 and MS COCO datasets.

Improves accuracy of image labeling on unseen data.

Effective in semi-supervised noisy label scenarios.

Abstract

Collecting large training datasets, annotated with high-quality labels, is costly and time-consuming. This paper proposes a novel framework for training deep convolutional neural networks from noisy labeled datasets that can be obtained cheaply. The problem is formulated using an undirected graphical model that represents the relationship between noisy and clean labels, trained in a semi-supervised setting. In our formulation, the inference over latent clean labels is tractable and is regularized during training using auxiliary sources of information. The proposed model is applied to the image labeling problem and is shown to be effective in labeling unseen images as well as reducing label noise in training on CIFAR-10 and MS COCO datasets.