KNN-enhanced Deep Learning Against Noisy Labels
This work tackles the problem of noisy labels for deep learning practitioners, offering an incremental improvement in label correction methods.
This paper addresses the challenge of noisy labels in deep learning by proposing an iterative method that combines deep neural networks for feature extraction with K-Nearest Neighbors for label inference. The approach demonstrates improved accuracy over existing label correction methods, achieving 76.78% on the Clothing1M dataset.
Supervised learning on Deep Neural Networks (DNNs) is data hungry. Optimizing performance of DNN in the presence of noisy labels has become of paramount importance since collecting a large dataset will usually bring in noisy labels. Inspired by the robustness of K-Nearest Neighbors (KNN) against data noise, in this work, we propose to apply deep KNN for label cleanup. Our approach leverages DNNs for feature extraction and KNN for ground-truth label inference. We iteratively train the neural network and update labels to simultaneously proceed towards higher label recovery rate and better classification performance. Experiment results show that under the same setting, our approach outperforms existing label correction methods and achieves better accuracy on multiple datasets, e.g.,76.78% on Clothing1M dataset.