Enhancing Robustness in Learning with Noisy Labels: An Asymmetric Co-Training Approach

Mengmeng Sheng, Zeren Sun, Gensheng Pei, Tao Chen, Haonan Luo, Yazhou Yao

Abstract:

Label noise, an inevitable issue in various real-world datasets, tends to impair the performance of deep neural networks.
A large body of literature focuses on symmetric co-training, aiming to enhance model robustness by exploiting interactions between models with distinct capabilities. However, the symmetric training processes employed in existing methods often culminate in model consensus, diminishing their efficacy in handling noisy labels. To this end, we propose an Asymmetric Co-Training (ACT) method to mitigate the detrimental effects of label noise. Specifically, we introduce an asymmetric training framework in which one model (i.e., RTM) is robustly trained with a selected subset of clean samples while the other (i.e., NTM) is conventionally trained using the entire training set. We propose two novel criteria based on agreement and discrepancy between models, establishing asymmetric sample selection and mining.
Moreover, a metric, derived from the divergence between models, is devised to quantify label memorization, guiding our method in determining the optimal stopping point for sample mining. Finally, we propose to dynamically re-weight identified clean samples according to their reliability inferred from historical information. We additionally employ consistency regularization to achieve further performance improvement. Extensive experimental results on synthetic and real-world datasets demonstrate the effectiveness and superiority of our method.

Primary Subject Area: [Content] Vision and Language

Secondary Subject Area: [Experience] Multimedia Applications

Relevance To Conference: This paper addresses the challenge of learning with noisy labels, which arise from errors in manual labeling or the collection of webly data for supervised learning. In this work, we introduce a novel asymmetric co-training approach to alleviate the harmful effects of noisy labels. Additionally, we bridge the gap between the two modalities of text and images in this paper. Furthermore, the label noise learning method we propose holds potential inspirational significance for cross-modal retrieval tasks with noisy labels. We believe our work will provide an alternative solution for the field and be of high interest to the ACM readership since it introduces a unique perspective on learning with noisy labels, delivering superior performance and efficiency. Notably, our proposed method surpasses existing state-of-the-art methods on various benchmark datasets, including synthetic and real-world datasets.

Submission Number: 572