Suppressing Mislabeled Data via Grouping and Self-Attention

TL;DR

This paper introduces Attentive Feature Mixup (AFM), a simple and efficient training method that improves noise robustness in deep learning by focusing on clean samples and suppressing mislabeled data through group-based attention and mixup interpolation.

Contribution

The paper proposes AFM, a novel plug-and-play module that constructs groups, assigns attention weights, and performs mixup to reduce the impact of noisy labels without extra clean data or assumptions.

Findings

Achieves state-of-the-art results on Food101N and Clothing1M datasets.

Effectively suppresses influence of mislabeled data via attention-weighted mixup.

Reduces the ratio of useless samples compared to original noisy datasets.

Abstract

Deep networks achieve excellent results on large-scale clean data but degrade significantly when learning from noisy labels. To suppressing the impact of mislabeled data, this paper proposes a conceptually simple yet efficient training block, termed as Attentive Feature Mixup (AFM), which allows paying more attention to clean samples and less to mislabeled ones via sample interactions in small groups. Specifically, this plug-and-play AFM first leverages a \textit{group-to-attend} module to construct groups and assign attention weights for group-wise samples, and then uses a \textit{mixup} module with the attention weights to interpolate massive noisy-suppressed samples. The AFM has several appealing benefits for noise-robust deep learning. (i) It does not rely on any assumptions and extra clean subset. (ii) With massive interpolations, the ratio of useless samples is reduced dramatically compared to the original noisy ratio. (iii) \pxj{It jointly optimizes the interpolation weights with classifiers, suppressing the influence of mislabeled data via low attention weights. (iv) It partially inherits the vicinal risk minimization of mixup to alleviate over-fitting while improves it by sampling fewer feature-target vectors around mislabeled data from the mixup vicinal distribution.} Extensive experiments demonstrate that AFM yields state-of-the-art results on two challenging real-world noisy datasets: Food101N and Clothing1M. The code will be available at https://github.com/kaiwang960112/AFM.