Robust Inference via Generative Classifiers for Handling Noisy Labels

TL;DR

This paper introduces RoG, a robust generative classifier that enhances neural network performance on noisy datasets by leveraging feature space modeling, without retraining the original model.

Contribution

The paper proposes a novel inference method, RoG, which constructs a generative classifier on pre-trained DNN features to improve robustness against noisy labels.

Findings

RoG improves classification accuracy on noisy datasets.

Ensemble RoG further enhances robustness.

Theoretical proof of RoG's superior generalization under noise.

Abstract

Large-scale datasets may contain significant proportions of noisy (incorrect) class labels, and it is well-known that modern deep neural networks (DNNs) poorly generalize from such noisy training datasets. To mitigate the issue, we propose a novel inference method, termed Robust Generative classifier (RoG), applicable to any discriminative (e.g., softmax) neural classifier pre-trained on noisy datasets. In particular, we induce a generative classifier on top of hidden feature spaces of the pre-trained DNNs, for obtaining a more robust decision boundary. By estimating the parameters of generative classifier using the minimum covariance determinant estimator, we significantly improve the classification accuracy with neither re-training of the deep model nor changing its architectures. With the assumption of Gaussian distribution for features, we prove that RoG generalizes better than baselines under noisy labels. Finally, we propose the ensemble version of RoG to improve its performance by investigating the layer-wise characteristics of DNNs. Our extensive experimental results demonstrate the superiority of RoG given different learning models optimized by several training techniques to handle diverse scenarios of noisy labels.