CUE: Concept-Aware Multi-Label Expansion to Mitigate Concept Confusion in Long-Tailed Learning
Ruichi Zhang Chikai Shang Jiacheng Yang Mengke Li, Yang Zhou Junlong Gao Yang Lu · 2026
Read CUE: Concept-Aware Multi-Label Expansion to Mitigate Concept Confusion in Long-Tailed Learning by Ruichi Zhang Chikai Shang Jiacheng Yang Mengke Li, Yang Zhou Junlong Gao Yang Lu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-01309.txt.
Description
Ruichi Zhang and six co-authors tackle a flaw in image training data where a few common classes have many examples and most rare classes have very few. Fine-tuning large pre-trained models helps, but training with one label per image can hide links between related classes and hurt rare ones.
Their method, called CUE, adds extra concept labels to each image. One set of cues comes from a vision model that reads the image itself, and a second set comes from a language model that describes each class. Two weighted loss terms teach the model these links while the main loss still sorts the classes. On several standard long-tailed test sets, CUE beats recent methods and gives rare and common classes a more even score.
Catalog record
| CREATOR | Ruichi Zhang Chikai Shang Jiacheng Yang Mengke Li, Yang Zhou Junlong Gao Yang Lu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.01309