When Optimizing $f$-divergence is Robust with Label Noise

Abstract

We show when maximizing a properly defined $f$-divergence measure with respect to a classifier's predictions and the supervised labels is robust with label noise. Leveraging its variational form, we derive a nice decoupling property for a family of $f$-divergence measures when label noise presents, where the divergence is shown to be a linear combination of the variational difference defined on the clean distribution and a bias term introduced due to the noise. The above derivation helps us analyze the robustness of different $f$-divergence functions. With established robustness, this family of $f$-divergence functions arises as useful metrics for the problem of learning with noisy labels, which do not require the specification of the labels' noise rate. When they are possibly not robust, we propose fixes to make them so. In addition to the analytical results, we present thorough experimental evidence. Our code is available at https://github.com/UCSC-REAL/Robust-f-divergence-measures.

arXiv commentPublished as a conference paper at ICLR 2021

References in corpus (10)

• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Classification with Noisy Labels by Importance Reweighting
• Training Convolutional Networks with Noisy Labels
• Noise Tolerance under Risk Minimization
• Making Risk Minimization Tolerant to Label Noise
• Mixture Proportion Estimation via Kernel Embedding of Distributions
• On Symmetric Losses for Learning from Corrupted Labels
• On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data
• Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels
• Does label smoothing mitigate label noise?

Cited by in corpus (2)

• Extended T: Learning with Mixed Closed-set and Open-set Noisy Labels
• A Symmetric Loss Perspective of Reliable Machine Learning