Curriculum Meta-Learning for Few-Shot Node Classification in Heterogeneous Networks
International Journal of Computational and Biological Sciences · 15.05.2026 · 3 (2) · 34-45
DOI: 10.66238/ijcbs80 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computational and Biological Sciences, online erschienen am 15.05.2026. Verfasst von Takumi Sasaki. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The rapid proliferation of complex, interconnected data across diverse domains has elevated the importance of heterogeneous networks, which encapsulate multiple types of nodes and edges. While traditional graph neural networks have demonstrated remarkable efficacy in node classification tasks, they fundamentally rely on massive quantities of labeled data. In many real-world scenarios, such as emerging disease detection or newly registered user classification, acquiring extensive annotations is prohibitively expensive and time-consuming. Meta-learning has emerged as a promising paradigm to address this few-shot learning problem by leveraging knowledge from seen classes to generalize to unseen classes with minimal examples. However, the extreme structural complexity of heterogeneous networks often causes standard meta-learning approaches to suffer from severe gradient confusion and sub-optimal convergence, particularly when faced with randomly sampled, highly difficult tasks early in the training process. To overcome these critical limitations, this paper proposes a novel framework that integrates curriculum learning with meta-learning for few-shot node classification in heterogeneous networks. By dynamically quantifying the difficulty of few-shot tasks based on structural node properties and feature space distinctiveness, the proposed framework strategically orders the training episodes from easy to hard. This methodical pacing guides the meta-learner through a smoothed optimization landscape, enabling more robust feature representations and superior generalization. Extensive conceptual analysis and simulated experimental evaluations demonstrate that the proposed curriculum meta-learning approach significantly mitigates the instability of episodic training, yielding substantial performance improvements over existing state-of-the-art baselines.
Zitieren
APA
Sasaki, T. (2026). Curriculum Meta-Learning for Few-Shot Node Classification in Heterogeneous Networks. International Journal of Computational and Biological Sciences, 3(2), 34-45. https://doi.org/10.66238/ijcbs80
Vancouver
Sasaki T. Curriculum Meta-Learning for Few-Shot Node Classification in Heterogeneous Networks. International Journal of Computational and Biological Sciences. 2026;3(2):34-45. doi:10.66238/ijcbs80
BibTeX
@article{Sasaki2026,
author = {Sasaki, Takumi},
title = {Curriculum Meta-Learning for Few-Shot Node Classification in Heterogeneous Networks},
journal = {International Journal of Computational and Biological Sciences},
year = {2026},
volume = {3},
number = {2},
pages = {34--45},
doi = {10.66238/ijcbs80}
}
EUVIMED