상세 보기
Federated Deep Reinforcement Learning for MEC-UAV Network: A Survey
- Mac, Huy Dang;
- Tran, Kiet Nguyen Tuan;
- Do, Tung Son;
- Cho, Sungrae
SCOPUS
0초록
The exponential proliferation of Internet of Things (IoT) devices has catalyzed an unprecedented demand for localized, low-latency computational capabilities. Unmanned Aerial Vehicles (UAVs) equipped with Mobile Edge Computing (MEC) servers present an agile, flexible solution to deliver computing resources directly to the network edge, particularly in infrastructure-deprived, disaster-stricken, or highly dynamic environments. However, optimizing the joint parameters of 3D trajectory design, communication resource allocation, and computation task offloading in such environments constitutes a highly complex, non-convex, and NP-hard problem. While centralized Deep Reinforcement Learning (DRL) algorithms have demonstrated remarkable efficacy in resolving these optimizations, they inherit severe limitations regarding high communication overhead, strict data centralization requirements, and associated user privacy vulnerabilities. Consequently, Federated Deep Reinforcement Learning (FDRL) has emerged as a disruptive paradigm that amalgamates the robust sequential decision-making capabilities of DRL with the privacy-preserving, distributed architecture of Federated Learning. This paper provides a comprehensive survey of the state-of-the-art architectures, mathematical models, and optimization methodologies characterizing FDRL in UAVassisted MEC networks. We systematically review the fundamental system models, analyze the algorithmic integration of FL and DRL, evaluate the impact on network efficiency, and extensively discuss the open challenges and future research directions shaping this nascent yet pivotal domain.
키워드
- 제목
- Federated Deep Reinforcement Learning for MEC-UAV Network: A Survey
- 저자
- Mac, Huy Dang; Tran, Kiet Nguyen Tuan; Do, Tung Son; Cho, Sungrae
- 발행일
- 2026-07
- 유형
- Conference Paper
- 페이지
- 122 ~ 126
- 언어
- ENG
- 출판사
- IEEE Computer Society
- 발행국가
- 미국
- 분량
- 5 페이지
- ISSN
- E 2165-8536
P 2165-8528