Argus: Federated Non-convex Bilevel Learning over 6G Space-Air-Ground Integrated Network

2025-05-14

AI Summary

To address infrastructure scarcity, time-varying topologies, and asynchronous delays in 6G space-air-ground integrated networks (SAGIN), this paper proposes Argus—the first asynchronous distributed federated bilevel learning framework tailored for nonconvex and nonsmooth scenarios. Argus departs from conventional synchronous assumptions by jointly integrating asynchronous optimization, bilevel programming, and dynamic topology modeling, while theoretically guaranteeing convergence and achieving superior iteration, communication, and computational complexity bounds. Experiments in dynamic SAGIN environments demonstrate that Argus accelerates training by 37% and reduces communication overhead by 29% compared to baseline methods. It effectively mitigates the straggler effect and significantly enhances both the robustness and efficiency of collaborative learning among mobile intelligent agents—such as unmanned aerial vehicles—under highly heterogeneous and volatile network conditions.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Multiagent Systems: Adversarial Agents
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The space-air-ground integrated network (SAGIN) has recently emerged as a core element in the 6G networks. However, traditional centralized and synchronous optimization algorithms are unsuitable for SAGIN due to infrastructureless and time-varying environments. This paper aims to develop a novel Asynchronous algorithm a.k.a. Argus for tackling non-convex and non-smooth decentralized federated bilevel learning over SAGIN. The proposed algorithm allows networked agents (e.g. autonomous aerial vehicles) to tackle bilevel learning problems in time-varying networks asynchronously, thereby averting stragglers from impeding the overall training speed. We provide a theoretical analysis of the iteration complexity, communication complexity, and computational complexity of Argus. Its effectiveness is further demonstrated through numerical experiments.

Problem

• Develops Argus for federated bilevel learning in 6G SAGIN
• Enables asynchronous learning in time-varying decentralized networks
• Addresses non-convex non-smooth optimization in infrastructureless environments

Innovation

• Asynchronous algorithm for federated bilevel learning
• Decentralized optimization in time-varying SAGIN networks
• Theoretical analysis of complexity for Argus