[Paper Review] Privacy-Preserving Federated Learning for UAV-Enabled Networks: Learning-Based Joint Scheduling and Resource Management

Helin Yang, Jun Zhao+4 more|arXiv (Cornell University)|Nov 28, 2020

Privacy-Preserving Technologies in Data29 references14 citations

TL;DR

This paper proposes an asynchronous federated learning (AFL) framework for UAV-enabled networks that preserves privacy by training models locally on devices without transmitting raw data. It employs an A3C-based algorithm for joint device selection, UAV placement, and resource management, achieving faster convergence and higher accuracy than existing methods in dynamic, resource-constrained environments.

ABSTRACT

Unmanned aerial vehicles (UAVs) are capable of serving as flying base stations (BSs) for supporting data collection, artificial intelligence (AI) model training, and wireless communications. However, due to the privacy concerns of devices and limited computation or communication resource of UAVs, it is impractical to send raw data of devices to UAV servers for model training. Moreover, due to the dynamic channel condition and heterogeneous computing capacity of devices in UAV-enabled networks, the reliability and efficiency of data sharing require to be further improved. In this paper, we develop an asynchronous federated learning (AFL) framework for multi-UAV-enabled networks, which can provide asynchronous distributed computing by enabling model training locally without transmitting raw sensitive data to UAV servers. The device selection strategy is also introduced into the AFL framework to keep the low-quality devices from affecting the learning efficiency and accuracy. Moreover, we propose an asynchronous advantage actor-critic (A3C) based joint device selection, UAVs placement, and resource management algorithm to enhance the federated convergence speed and accuracy. Simulation results demonstrate that our proposed framework and algorithm achieve higher learning accuracy and faster federated execution time compared to other existing solutions.

Motivation & Objective

• To address privacy concerns in UAV-enabled networks by avoiding transmission of raw device data to UAV servers.
• To improve learning efficiency and convergence speed in dynamic, resource-constrained UAV networks with heterogeneous devices.
• To jointly optimize device selection, UAV placement, and resource allocation to minimize federated learning execution time.
• To enhance reliability and robustness under varying channel conditions and device mobility.

Proposed method

• Develops an asynchronous federated learning (AFL) framework that enables local model training on devices, eliminating the need to send raw data to UAV servers.
• Introduces a device selection strategy to exclude low-quality devices and maintain model accuracy and training efficiency.
• Proposes an asynchronous advantage actor-critic (A3C) reinforcement learning algorithm to jointly optimize device selection, UAV positioning, and resource allocation.
• Uses linear function approximation for the action-value function $ Q(s,a;\bm{\theta}_c) = \bm{\theta}_c^T \cdot \bm{\Psi}(s,a) $ to estimate state-action values.
• Employs a critic network that minimizes the Mean Square Projected Bellman Error to approximate the optimal action-value function.
• Coordinates multiple agents in a multi-UAV setting, enabling each to update a shared estimation of the global action-value function for joint decision-making.

Experimental results

Research questions

• RQ1 How can privacy be preserved in UAV-enabled networks when training AI models across mobile devices?
• RQ2 What is the impact of device heterogeneity and dynamic channel conditions on federated learning convergence in UAV networks?
• RQ3 How can joint optimization of device selection, UAV placement, and resource allocation improve federated learning performance?
• RQ4 Can a multi-agent A3C approach effectively coordinate UAVs and devices to accelerate convergence and reduce execution time?

Key findings

• The proposed AFL framework achieves higher model accuracy compared to baseline methods due to effective device selection and privacy-preserving training.
• The A3C-based joint optimization algorithm reduces federated learning execution time by efficiently managing UAV placement and resource allocation.
• Simulation results show faster convergence speed under dynamic channel conditions and varying device mobility.
• The method maintains high learning accuracy even when low-quality devices are present, thanks to the device selection mechanism.
• The A3C algorithm converges to a stable solution with high probability, as proven by the convergence analysis of the actor-critic networks.