[Paper Review] Federated Learning for Task and Resource Allocation in Wireless High Altitude Balloon Networks

Sihua Wang, Mingzhe Chen+5 more|arXiv (Cornell University)|Mar 19, 2020

UAV Applications and Optimization35 references18 citations

TL;DR

This paper proposes a federated learning (FL)-based support vector machine (SVM) framework for dynamic user association, task partitioning, and resource allocation in high-altitude balloon (HAB)-enabled mobile edge computing (MEC) networks. By enabling HABs to collaboratively train an SVM model without sharing raw user data, the method proactively predicts optimal user associations, reducing the weighted sum of energy and delay by up to 16.1% compared to centralized methods.

ABSTRACT

In this paper, the problem of minimizing energy and time consumption for task computation and transmission is studied in a mobile edge computing (MEC)-enabled balloon network. In the considered network, each user needs to process a computational task in each time instant, where high-altitude balloons (HABs), acting as flying wireless base stations, can use their powerful computational abilities to process the tasks offloaded from their associated users. Since the data size of each user's computational task varies over time, the HABs must dynamically adjust the user association, service sequence, and task partition scheme to meet the users' needs. This problem is posed as an optimization problem whose goal is to minimize the energy and time consumption for task computing and transmission by adjusting the user association, service sequence, and task allocation scheme. To solve this problem, a support vector machine (SVM)-based federated learning (FL) algorithm is proposed to determine the user association proactively. The proposed SVM-based FL method enables each HAB to cooperatively build an SVM model that can determine all user associations without any transmissions of either user historical associations or computational tasks to other HABs. Given the prediction of the optimal user association, the service sequence and task allocation of each user can be optimized so as to minimize the weighted sum of the energy and time consumption. Simulations with real data of city cellular traffic from the OMNILab at Shanghai Jiao Tong University show that the proposed algorithm can reduce the weighted sum of the energy and time consumption of all users by up to 16.1% compared to a conventional centralized method.

Motivation & Objective

• To address dynamic task offloading and resource allocation in mobile edge computing (MEC)-enabled high-altitude balloon (HAB) networks with varying task sizes.
• To minimize the weighted sum of energy and time consumption for task computation and transmission across all users.
• To enable HABs to make proactive user association decisions without sharing sensitive user data or historical associations.
• To design a distributed, privacy-preserving optimization framework that avoids centralized data collection and reduces computational overhead.

Proposed method

• A support vector machine (SVM)-based federated learning (FL) algorithm is proposed to enable HABs to collaboratively train a global user association model without exchanging raw user data.
• Each HAB trains a local SVM model using its own user association history, and model parameters are aggregated via FL to build a global model for proactive user association prediction.
• The user association prediction is used as input to optimize the service sequence and task partitioning scheme for each HAB.
• The optimization problem is formulated as a mixed-integer non-linear program (MINLP), and a sequential convex approximation method is applied to solve it efficiently.
• The method ensures privacy by preventing transmission of user task data or historical associations to other HABs.
• Convergence and suboptimality bounds are theoretically analyzed using a Lyapunov-based approach, proving global convergence of the algorithm.

Experimental results

Research questions

• RQ1 Can federated learning be effectively used to enable HABs to predict optimal user associations without sharing sensitive user data or historical associations?
• RQ2 How can dynamic user association, service sequencing, and task partitioning be jointly optimized to minimize energy and delay in HAB-MEC networks with time-varying task sizes?
• RQ3 What is the performance gain of a decentralized, privacy-preserving FL-based approach compared to a conventional centralized optimization method?
• RQ4 How does the proposed method maintain low communication and computational overhead while ensuring convergence and optimality?

Key findings

• The proposed FL-based SVM method reduces the weighted sum of energy and time consumption across all users by up to 16.1% compared to a conventional centralized optimization method.
• The algorithm achieves convergence with a suboptimality gap that decreases exponentially over iterations, as proven by the theoretical analysis.
• The federated learning framework preserves user privacy by ensuring no raw task data or historical associations are shared between HABs.
• The method significantly reduces communication and computational overhead compared to traditional iterative optimization methods like Lagrangian dual decomposition.
• Simulations using real-world cellular traffic data from Shanghai Jiao Tong University demonstrate consistent performance gains across various network conditions.
• The proposed approach enables proactive user association decisions, reducing the delay associated with online optimization and data collection.