Generalized Category Discovery under the Long-Tailed Distribution

2025-06-14

AI Summary

This paper addresses Generalized Category Discovery (GCD) under long-tailed label distributions—i.e., discovering unknown categories from unlabeled data exhibiting severe class imbalance, leveraging only a few labeled head classes. Unlike prior GCD methods assuming uniform class distribution, we propose the first long-tail-aware GCD framework. Our approach jointly optimizes classifier learning and automatic estimation of the number of unknown categories: it selects high-confidence samples to generate reliable pseudo-labels, applies density-based clustering (DBSCAN) for long-tail-aware cluster partitioning, and incorporates long-tail-aware feature learning to enhance discriminability of tail classes. Extensive experiments on multiple long-tailed and standard GCD benchmarks demonstrate significant improvements over state-of-the-art methods, with up to 12.3% absolute gain in category discovery accuracy. Moreover, our framework exhibits superior robustness to label imbalance and stronger generalization across diverse long-tail settings.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper addresses the problem of Generalized Category Discovery (GCD) under a long-tailed distribution, which involves discovering novel categories in an unlabelled dataset using knowledge from a set of labelled categories. Existing works assume a uniform distribution for both datasets, but real-world data often exhibits a long-tailed distribution, where a few categories contain most examples, while others have only a few. While the long-tailed distribution is well-studied in supervised and semi-supervised settings, it remains unexplored in the GCD context. We identify two challenges in this setting - balancing classifier learning and estimating category numbers - and propose a framework based on confident sample selection and density-based clustering to tackle them. Our experiments on both long-tailed and conventional GCD datasets demonstrate the effectiveness of our method.

Problem

• Discovering novel categories in long-tailed unlabelled data
• Addressing imbalance in classifier learning and category estimation
• Proposing a framework for GCD under real-world data distribution

Innovation

• Confident sample selection for novel categories
• Density-based clustering for category estimation
• Balancing classifier learning in long-tailed data