GOAL: Geometrically Optimal Alignment for Continual Generalized Category Discovery

2026-02-23

AI Summary

This work addresses generalized category discovery in continual learning, where the goal is to identify novel classes from unlabeled data while mitigating catastrophic forgetting and feature misalignment for known classes. The authors propose GOAL, a unified framework that, for the first time, integrates a fixed Equiangular Tight Frame (ETF) classifier into this setting. By leveraging geometric constraints, GOAL establishes a stable feature space: supervised feature alignment is applied to known classes, while a confidence-guided self-alignment strategy is employed for novel classes, enabling synergistic learning across old and new categories. Evaluated on four benchmarks, GOAL outperforms the current state-of-the-art method, Happy, reducing the forgetting rate by 16.1% and improving novel class discovery accuracy by 3.2%.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• 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
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Continual Generalized Category Discovery (C-GCD) requires identifying novel classes from unlabeled data while retaining knowledge of known classes over time. Existing methods typically update classifier weights dynamically, resulting in forgetting and inconsistent feature alignment. We propose GOAL, a unified framework that introduces a fixed Equiangular Tight Frame (ETF) classifier to impose a consistent geometric structure throughout learning. GOAL conducts supervised alignment for labeled samples and confidence-guided alignment for novel samples, enabling stable integration of new classes without disrupting old ones. Experiments on four benchmarks show that GOAL outperforms the prior method Happy, reducing forgetting by 16.1% and boosting novel class discovery by 3.2%, establishing a strong solution for long-horizon continual discovery.

Problem

• Continual Generalized Category Discovery
• catastrophic forgetting
• feature alignment
• novel class discovery
• class-incremental learning

Innovation

• Equiangular Tight Frame
• Continual Generalized Category Discovery
• Feature Alignment
• Catastrophic Forgetting
• Confidence-guided Learning