When Domain Generalization meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach

2025-03-19

AI Summary

Generalized Category Discovery (GCD) faces critical challenges in realistic settings where target-domain data is unavailable and significant source-to-target distribution shifts exist. Method: This paper introduces Domain-Generalization GCD (DG-GCD), the first paradigm enabling GCD without access to any target-domain data. It learns domain-invariant, discriminative embedding spaces solely from source-domain data via an episodic fine-tuning mechanism based on adaptive weighted task arithmetic—integrating synthetic-domain augmentation, cross-domain GCD task sampling, and a novel margin-based contrastive loss. Additionally, it employs foundation-model-driven synthetic domain generation and open-set domain adaptation. Contribution/Results: Evaluated on three standard benchmarks under strict zero-shot target-domain data conditions, DG-GCD consistently outperforms existing GCD methods, achieving state-of-the-art clustering and classification performance. It establishes a generalizable framework for unsupervised cross-domain category discovery.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Generalized Class Discovery (GCD) clusters base and novel classes in a target domain using supervision from a source domain with only base classes. Current methods often falter with distribution shifts and typically require access to target data during training, which can sometimes be impractical. To address this issue, we introduce the novel paradigm of Domain Generalization in GCD (DG-GCD), where only source data is available for training, while the target domain, with a distinct data distribution, remains unseen until inference. To this end, our solution, DG2CD-Net, aims to construct a domain-independent, discriminative embedding space for GCD. The core innovation is an episodic training strategy that enhances cross-domain generalization by adapting a base model on tasks derived from source and synthetic domains generated by a foundation model. Each episode focuses on a cross-domain GCD task, diversifying task setups over episodes and combining open-set domain adaptation with a novel margin loss and representation learning for optimizing the feature space progressively. To capture the effects of fine-tuning on the base model, we extend task arithmetic by adaptively weighting the local task vectors concerning the fine-tuned models based on their GCD performance on a validation distribution. This episodic update mechanism boosts the adaptability of the base model to unseen targets. Experiments across three datasets confirm that DG2CD-Net outperforms existing GCD methods customized for DG-GCD.

Problem

• Addresses domain shifts in Generalized Class Discovery (GCD).
• Introduces DG-GCD for training with only source domain data.
• Proposes DG2CD-Net for domain-independent embedding space creation.

Innovation

• Episodic training enhances cross-domain generalization
• Adaptive task arithmetic optimizes model adaptability
• Domain-independent embedding space for class discovery