Prototype-Anchored Learning for Learning with Imperfect Annotations

TL;DR

This paper introduces prototype-anchored learning (PAL), a method that improves classification from imperfect annotations by anchoring class prototypes, tightening generalization bounds, and enhancing robustness to noise and imbalance.

Contribution

The paper proposes a novel prototype-anchored learning approach that leverages theoretical insights to improve learning with noisy, imbalanced, or biased annotations.

Findings

PAL improves robustness to label noise.

PAL enhances performance on imbalanced datasets.

Theoretical analysis supports the effectiveness of prototype anchoring.

Abstract

The success of deep neural networks greatly relies on the availability of large amounts of high-quality annotated data, which however are difficult or expensive to obtain. The resulting labels may be class imbalanced, noisy or human biased. It is challenging to learn unbiased classification models from imperfectly annotated datasets, on which we usually suffer from overfitting or underfitting. In this work, we thoroughly investigate the popular softmax loss and margin-based loss, and offer a feasible approach to tighten the generalization error bound by maximizing the minimal sample margin. We further derive the optimality condition for this purpose, which indicates how the class prototypes should be anchored. Motivated by theoretical analysis, we propose a simple yet effective method, namely prototype-anchored learning (PAL), which can be easily incorporated into various learning-based classification schemes to handle imperfect annotation. We verify the effectiveness of PAL on class-imbalanced learning and noise-tolerant learning by extensive experiments on synthetic and real-world datasets.