LANE: Label-Aware Noise Elimination for Fine-Grained Text Classification

Tiberiu Sosea, Cornelia Caragea

Keywords: label noise, training dynamics, example reweighting

TL;DR: Improve learning under label noise by utilizing training dynamics and the relations between classes.

Abstract: We propose Label-Aware Noise Elimination (LANE), a new approach that improves the robustness of deep learning models in fine-grained text classification when trained under increased label noise. LANE leverages the semantic relations between classes and monitors the training dynamics of the model on each training example to dynamically lower the importance of training examples that may have noisy labels. We test the effectiveness of LANE in fine-grained text classification and benchmark our approach on a wide variety of datasets with various number of classes and various amounts of label noise. LANE considerably outperforms strong baselines on all datasets, obtaining significant improvements ranging from an average improvement of 2.4% in F1 on manually annotated datasets to a considerable average improvement of 4.5% F1 on datasets with injected noisy labels. We carry out comprehensive analyses of LANE and identify the key components that lead to its success.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10296