Improving Detection of Rare Nodes in Hierarchical Multi-Label Learning

TMLR Paper6083 Authors

Abstract: In hierarchical multi-label classification, a persistent challenge is enabling model predictions to reach deeper levels of the hierarchy for more detailed or fine-grained classifications. This difficulty partly arises from the natural rarity of certain classes (or hierarchical nodes) and the hierarchical constraint that ensures child nodes are almost always less frequent than their parents. To address this, we propose a weighted loss objective for neural networks that combines node-wise imbalance weighting with focal weighting components, the latter leveraging modern quantification of ensemble uncertainties. By emphasizing rare nodes rather than rare observations (data points), and focusing on uncertain nodes for each model output distribution during training, we observe improvements in recall by up to a factor of five on benchmark datasets, along with statistically significant gains in F1 score. We also show our approach aids convolutional networks on challenging tasks, as in situations with suboptimal encoders or limited data.

Submission Length: Regular submission (no more than 12 pages of main content)

Previous TMLR Submission Url: https://openreview.net/forum?id=9T2nhQXFnM&noteId=aQnbGpdi6B

Changes Since Last Submission:

Due to an incorrect assumption in the paper standard being A4, the previous submission was desk-rejected. The offending setting has been removed, and the paper size is now the default US "Letter" size. Additionally, some minor text revisions have been performed to enhance clarity and terseness.

Additional revisions have been made throughout the work in the form of general editing, and added experiments, following requests, suggestions, and advice from reviewers.

Assigned Action Editor: ~Fernando_Perez-Cruz1

Submission Number: 6083