A meta-learning and self-supervised framework for fast convergence in large-scale data optimization
TL;DR
A Meta-SelfSupervised Convergence Optimization Framework (MSCOF), which fuses self-supervised representation learning with meta-learning strategies to enhance adaptability and reduce convergence time under limited supervision, and maintains robustness across diverse data distributions and network conditions.
Abstract
With the exponential growth of heterogeneous data across distributed environments, achieving rapid convergence and stable optimization remains a major challenge in large-scale machine learning. This paper introduces a Meta-SelfSupervised Convergence Optimization Framework (MSCOF), which fuses self-supervised representation learning with meta-learning strategies to enhance adaptability and reduce convergence time under limited supervision. The proposed framework dynamically adjusts gradient feedback through an adaptive controller and employs distributed optimization to achieve global stability with minimal synchronization cost. Experimental results on large-scale synthetic and real-world datasets demonstrate that MSCOF achieves up to 35% faster convergence and 25% lower gradient variance compared with state-of-the-art baselines. The model maintains robustness across diverse data distributions and network conditions, confirming its suitability for real-world big data processing. This study contributes a reproducible and scalable approach for accelerating learning efficiency in intelligent distributed systems.