Deep k-NN for Noisy Labels
arXiv:2004.12289
Abstract
Modern machine learning models are often trained on examples with noisy labels that hurt performance and are hard to identify. In this paper, we provide an empirical study showing that a simple -nearest neighbor-based filtering approach on the logit layer of a preliminary model can remove mislabeled training data and produce more accurate models than many recently proposed methods. We also provide new statistical guarantees into its efficacy.
Full paper (including supplemental) can be found at https://github.com/dbahri/deepknn
References in corpus (14)
- Learning to Reweight Examples for Robust Deep Learning
- Classification with Noisy Labels by Importance Reweighting
- MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels
- Training Convolutional Networks with Noisy Labels
- Deep Learning is Robust to Massive Label Noise
- Deep k-Nearest Neighbors: Towards Confident, Interpretable and Robust Deep Learning
- Fast learning rates for plug-in classifiers
- Learning From Noisy Singly-labeled Data
- Combating Label Noise in Deep Learning Using Abstention
- Rates of Convergence for Nearest Neighbor Classification
- Robust Inference via Generative Classifiers for Handling Noisy Labels
- Learning From Noisy Large-Scale Datasets With Minimal Supervision
- On the Resistance of Nearest Neighbor to Random Noisy Labels
- Fast Rates for a kNN Classifier Robust to Unknown Asymmetric Label Noise