Robust Losses from Univariate Base Functions for Noisy-Label Learning
Machine Learning
Summary
The authors address the problem of training deep neural networks when the labels used for training are incorrect or noisy. They introduce a new general method to create loss functions that are more robust to noise by building them from simpler single-variable functions. Their framework includes two methods called Target Separation and Binary Reduction, each with theoretical conditions to ensure robustness against label noise. Experiments show that their approach works well across different types of noisy data. This work provides a clearer way to design and analyze loss functions that can handle noisy labels in multiclass problems.
Authors
Peng Hu, Jianwei Ma
Abstract
Learning with noisy labels is a fundamental problem in training reliable deep neural networks. Robust loss functions provide a direct and effective way to mitigate the adverse effects of label noise. However, most existing robust losses are designed directly at the level of the final multiclass objective, which makes it difficult to systematically characterize and extend their robustness properties. In this paper, we propose a general framework that constructs robust multiclass losses from univariate base functions. By defining mapping operators from base functions to multiclass losses, the robustness of the induced losses can be characterized through simple properties of the base functions. We develop two complementary construction schemes, Target Separation and Binary Reduction, corresponding to inter-class independent and inter-class dependent formulations, respectively. For both schemes, we analyze their symmetry and asymmetry properties and derive corresponding sufficient conditions, which provide theoretical criteria for noise-robust loss design. The proposed framework also provides a new route to constructing symmetric losses, serving as a complement to normalization-based symmetric loss designs. Extensive experiments on synthetic and real-world noisy-label benchmarks demonstrate that the proposed losses achieve competitive or superior performance under various noise settings.