CUE: Concept-Aware Multi-Label Expansion to Mitigate Concept Confusion in Long-Tailed Learning

2026-05-02

AI Summary

This work addresses the challenge of concept confusion in long-tailed learning caused by the strict single-label exclusivity assumption, which hinders feature sharing and discriminability among semantically related classes. To mitigate this issue, the paper introduces a multi-label formulation into long-tailed learning for the first time. It leverages CLIP to extract instance-level visual features and employs a large language model to generate class-level semantic priors. A unified framework combining Binary Logit Adjustment (BLA) and Logit Adjustment (LA) with adaptive weighting is proposed to refine the classifier. This approach effectively restores disrupted inter-class semantic relationships and substantially improves performance on tail classes, achieving state-of-the-art and well-balanced results across multiple long-tailed benchmarks.

Abstract

Long-tailed distributions are common in real-world recognition tasks, where a few head classes have many samples while most tail classes have very few. Recently, fine-tuning foundation models for long-tailed learning has gained attention due to their excellent performance. However, most existing methods focus solely on mitigating long-tailed distribution bias while overlooking concept confusion caused by the long-tailed distribution. In this paper, we study this problem and attribute it to the mutual exclusivity of single-label supervision under long-tailed distributions, which suppresses feature sharing among related classes and amplifies the dominance of head classes, leading to disrupted inter-class discriminability. To address this, we propose CUE, Concept-aware mUlti-label Expansion, which introduces multi-label concept signals to preserve disrupted inter-class relationships. Specifically, CUE constructs concept sets by (i) extracting instance-level visual cues from zero-shot CLIP and (ii) generating class-level semantic cues with LLM; the two cues are incorporated via separately weighted Binary Logit-Adjustment (BLA) auxiliary losses and jointly optimized with the baseline Logit-Adjustment (LA) loss. Experiments on several long-tailed benchmarks, CUE achieves balanced and strong performance, surpassing recent state-of-the-art methods. Code is available at: https://github.com/zhangruichi/CUE.

Problem

• long-tailed learning
• concept confusion
• single-label supervision
• inter-class discriminability
• feature sharing

Innovation

• long-tailed learning
• concept confusion
• multi-label expansion
• foundation models
• logit adjustment