Generalized Category Discovery via Reciprocal Learning and Class-Wise Distribution Regularization

2025-06-03

AI Summary

Generalized Category Discovery (GCD) aims to jointly identify both known and unknown classes from unlabeled data containing instances of both. However, existing parametric approaches suffer from degraded discriminability on base classes due to reliance on unreliable self-supervised signals. This paper proposes the Reciprocal Learning Framework (RLF), which jointly refines features and pseudo-labels via mutual feedback between a backbone network and an auxiliary branch. To mitigate base-class bias and enhance confidence in novel classes, we further introduce Class-level Distribution Regularization (CDR), which enforces consistency of class probability distributions using KL divergence. Our method integrates soft-label feedback and pseudo-label distillation without additional computational overhead. Evaluated on seven GCD benchmarks, RLF achieves state-of-the-art performance: it significantly improves novel-class accuracy while preserving base-class discrimination—demonstrating effective joint learning without performance trade-offs.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Generalized Category Discovery (GCD) aims to identify unlabeled samples by leveraging the base knowledge from labeled ones, where the unlabeled set consists of both base and novel classes. Since clustering methods are time-consuming at inference, parametric-based approaches have become more popular. However, recent parametric-based methods suffer from inferior base discrimination due to unreliable self-supervision. To address this issue, we propose a Reciprocal Learning Framework (RLF) that introduces an auxiliary branch devoted to base classification. During training, the main branch filters the pseudo-base samples to the auxiliary branch. In response, the auxiliary branch provides more reliable soft labels for the main branch, leading to a virtuous cycle. Furthermore, we introduce Class-wise Distribution Regularization (CDR) to mitigate the learning bias towards base classes. CDR essentially increases the prediction confidence of the unlabeled data and boosts the novel class performance. Combined with both components, our proposed method, RLCD, achieves superior performance in all classes with negligible extra computation. Comprehensive experiments across seven GCD datasets validate its superiority. Our codes are available at https://github.com/APORduo/RLCD.

Problem

• Identify unlabeled samples using labeled data knowledge
• Improve base discrimination in parametric-based GCD methods
• Mitigate learning bias towards base classes in GCD

Innovation

• Reciprocal Learning Framework enhances base classification
• Class-wise Distribution Regularization reduces learning bias
• Combined RLCD method improves all class performances