M2AML: Metric-Based Model-Agnostic Meta-Learning for Few-Shot Classification
Entropy · 23.04.2026 · 28 (5) · 484
DOI: 10.3390/e28050484 ↗Kurz gefasst
Fachartikel in der Zeitschrift Entropy, online erschienen am 23.04.2026. Verfasst von Xiaoming Han und 3 weiteren Autor:innen. Beteiligte Einrichtung: College of Computer Science and Technology, National University of Defense Technology, Changsha 410000, China. Die Arbeit nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Model-Agnostic Meta-Learning (MAML) and Prototypical Networks (ProtoNet) establish the foundational paradigms for few-shot classification. However, MAML suffers from optimization instability caused by reconstructing classification boundaries for every new task. Conversely, ProtoNet lacks the internal mathematical capacity necessary for task-specific parameter adaptation under domain shifts. To reconcile these structural limitations, we introduce Metric-based Model-Agnostic Meta-Learning (M2AML). By completely excising the parameterized classification layer from the episodic adaptation sequence, our framework replaces traditional inner-loop classification with a dynamic self-exclusive geometric similarity metric. Substituting functional mappings with spatial distance optimizations efficiently resolves evaluation conflicts, thereby establishing perfectly synchronized inner and outer learning rates alongside substantially accelerated adaptation steps. Extensive experiments across mini-ImageNet, tiered-ImageNet, and CIFAR-FS validate our approach against a comprehensive array of established algorithms. To ensure strictly fair comparative evaluations, we meticulously reproduce the MAML, ProtoNet, and Proto-MAML baselines. Empirical results demonstrate that M2AML achieves state-of-the-art performance across most evaluation settings, delivering absolute accuracy improvements ranging from 0.1% to 2.1% over existing leading models.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- College of Computer Science and Technology, National University of Defense Technology, Changsha 410000, China.
- School of Electronics and Information, Xi'an Polytechnic University, Xi'an 710048, China.
ORCID-Profile: Xiaoming Han, Zhen Wang
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Han, X., Shi, D., Wang, Z., & Yang, S. (2026). M2AML: Metric-Based Model-Agnostic Meta-Learning for Few-Shot Classification. Entropy, 28(5), 484. https://doi.org/10.3390/e28050484
Vancouver
Han X, Shi D, Wang Z, Yang S. M2AML: Metric-Based Model-Agnostic Meta-Learning for Few-Shot Classification. Entropy. 2026;28(5):484. doi:10.3390/e28050484
BibTeX
@article{Han2026,
author = {Han, Xiaoming and Shi, Dianxi and Wang, Zhen and Yang, Shaowu},
title = {M2AML: Metric-Based Model-Agnostic Meta-Learning for Few-Shot Classification},
journal = {Entropy},
year = {2026},
volume = {28},
number = {5},
pages = {484},
doi = {10.3390/e28050484}
}
EUVIMED