Unlocking the Potential of Unlabeled Data in Semi-Supervised Domain Generalization

2025-03-18

AI Summary

Semi-supervised domain generalization (SSDG) confronts challenges including distribution shifts between training and test domains, scarcity of labeled data, and absence of domain annotations. Existing methods leverage only high-confidence unlabeled samples while discarding low-confidence ones—despite their potential discriminative value. This work is the first to systematically exploit discriminative information embedded in “low-confidence unlabeled samples.” We propose a plug-and-play framework that requires no domain labels: (i) the UPC module employs proxy-based contrastive learning to enhance cross-domain feature consistency; (ii) the SC module implements proxy-based class learning to model the latent class structure among low-confidence samples; and (iii) we integrate confusion-class set construction with semi-supervised consistency regularization. Evaluated on four mainstream SSDG benchmarks, our method significantly outperforms state-of-the-art approaches, improving class discrimination and effectively mitigating inter-domain discrepancies.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Application Domains: Humanities & Computational Social Science
• Computer Vision: Diffusion Models for Vision

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

We address the problem of semi-supervised domain generalization (SSDG), where the distributions of train and test data differ, and only a small amount of labeled data along with a larger amount of unlabeled data are available during training. Existing SSDG methods that leverage only the unlabeled samples for which the model's predictions are highly confident (confident-unlabeled samples), limit the full utilization of the available unlabeled data. To the best of our knowledge, we are the first to explore a method for incorporating the unconfident-unlabeled samples that were previously disregarded in SSDG setting. To this end, we propose UPCSC to utilize these unconfident-unlabeled samples in SSDG that consists of two modules: 1) Unlabeled Proxy-based Contrastive learning (UPC) module, treating unconfident-unlabeled samples as additional negative pairs and 2) Surrogate Class learning (SC) module, generating positive pairs for unconfident-unlabeled samples using their confusing class set. These modules are plug-and-play and do not require any domain labels, which can be easily integrated into existing approaches. Experiments on four widely used SSDG benchmarks demonstrate that our approach consistently improves performance when attached to baselines and outperforms competing plug-and-play methods. We also analyze the role of our method in SSDG, showing that it enhances class-level discriminability and mitigates domain gaps. The code is available at https://github.com/dongkwani/UPCSC.

Problem

• Improves semi-supervised domain generalization with unlabeled data
• Incorporates unconfident-unlabeled samples previously ignored in SSDG
• Enhances class-level discriminability and reduces domain gaps

Innovation

• Utilizes unconfident-unlabeled samples in SSDG
• Introduces UPC and SC modules for learning
• Enhances class-level discriminability and domain generalization