Prototype-Anchored Learning for Learning with Imperfect Annotations
It addresses the problem of training deep neural networks with imperfect annotations for researchers and practitioners, but it is incremental as it builds on existing loss functions and classification schemes.
The paper tackles learning unbiased classification models from datasets with imperfect annotations like class imbalance or label noise by proposing prototype-anchored learning (PAL), which tightens generalization error bounds and shows effectiveness in experiments on synthetic and real-world datasets.
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.