[Paper Review] A VHetNet-Enabled Asynchronous Federated Learning-Based Anomaly Detection Framework for Ubiquitous IoT

Weili Wang, Omid Abbasi+4 more|arXiv (Cornell University)|Mar 6, 2023

UAV Applications and Optimization4 citations

TL;DR

This paper proposes a VHetNet-enabled asynchronous federated learning (AFL) framework that leverages high-altitude platform stations (HAPS) and UAVs to collaboratively train a global anomaly detection model for ubiquitous IoT, using differential privacy and a CA2C-based joint optimization of device association, UAV selection, and trajectory planning. The approach achieves high detection accuracy, low energy consumption, and reduced training time compared to baseline methods.

ABSTRACT

Anomaly detection for the Internet of Things (IoT) is a major intelligent service required by many fields, including intrusion detection, device-activity analysis, and security supervision. However, the heterogeneous distribution of data and resource-constrained end nodes present challenges for existing anomaly detection models. Due to the advantages of flexible deployment and multi-dimensional resources, high altitude platform stations (HAPSs) and unmanned aerial vehicles (UAVs), which are important components of vertical heterogeneous networks (VHetNets), have significant potential for sensing, computing, storage, and communication applications in ubiquitous IoT systems. In this paper, we propose a novel VHetNet-enabled asynchronous federated learning (AFL) framework to enable decentralized UAVs to collaboratively train a global anomaly detection model. In the VHetNet-enabled AFL framework, a HAPS operates as a central aerial server, and the local models trained in UAVs are uploaded to the HAPS for global aggregation due to its wide coverage and strong storage and computation capabilities. We introduce a UAV selection strategy into the AFL framework to prevent UAVs with low local model quality and large energy consumption from affecting the learning efficiency and detection accuracy of the global model. To ensure the security of transmissions between UAVs and the HAPS, we add designed noise to local model parameters in UAVs to achieve differential privacy. Moreover, we propose a compound-action actor-critic (CA2C)-based joint device association, UAV selection, and UAV trajectory planning algorithm to further enhance the overall federated execution efficiency and detection model accuracy. Extensive experimental evaluation on a real-world dataset demonstrates that the proposed algorithm can achieve high detection accuracy with short federated execution time and low energy consumption.

Motivation & Objective

• To address the challenges of data heterogeneity and resource constraints in ubiquitous IoT systems for anomaly detection.
• To enable decentralized, privacy-preserving model training across UAVs using asynchronous federated learning (AFL).
• To enhance learning efficiency and detection accuracy by integrating UAV trajectory planning, device association, and UAV selection into a unified optimization framework.
• To ensure secure model parameter exchange between UAVs and HAPS using differentially private WGAN-GP.
• To minimize energy consumption and training delay in UAV-assisted anomaly detection under limited onboard energy.

Proposed method

• A VHetNet architecture is employed, with HAPS as a central aerial server and UAVs as mobile edge nodes for local model training.
• Asynchronous Federated Learning (AFL) is adopted to allow partial UAV participation per round, reducing training delay and communication overhead.
• A UAV selection strategy is introduced to exclude low-quality or high-energy-consuming UAVs, improving global model convergence and accuracy.
• Differential privacy is enforced by adding calibrated noise to local model parameters before transmission to the HAPS, ensuring data privacy.
• A compound-action actor-critic (CA2C) reinforcement learning algorithm jointly optimizes device association, UAV selection, and UAV trajectory planning under energy constraints.
• A differentially private WGAN-GP is used as the local anomaly detection model to enhance privacy and detection performance.

Experimental results

Research questions

• RQ1 How can a decentralized anomaly detection framework be designed to handle data heterogeneity and resource constraints in ubiquitous IoT?
• RQ2 What is the impact of asynchronous federated learning with UAV selection on model convergence and detection accuracy in UAV-assisted IoT systems?
• RQ3 How can differential privacy be effectively integrated into the model aggregation process to protect sensitive data in UAV-HAPS communications?
• RQ4 To what extent can joint optimization of device association, UAV selection, and trajectory planning improve energy efficiency and learning efficiency in UAV-enabled federated learning?
• RQ5 How does the proposed CA2C-based scheduling strategy compare to baseline methods in terms of detection accuracy, energy consumption, and training time?

Key findings

• The proposed CA2C-AFL algorithm achieves the highest detection accuracy on unseen test data, outperforming DDPG-FL, DQN-AFL, and standalone models.
• The average UAV energy consumption of the CA2C-AFL algorithm is the lowest among all compared algorithms, due to optimized trajectory planning and selective model updates.
• The CA2C-AFL framework reduces both average and maximum training execution times, outperforming DDPG-FL and standalone methods due to asynchronous learning and efficient UAV subset selection.
• The detection accuracy of the standalone algorithm drops significantly on new datasets, indicating poor generalization due to lack of data diversity and model exchange.
• The DDPG-FL algorithm has the longest training time because it requires all UAVs to complete local updates before global aggregation, violating the asynchronous principle.
• The integration of differential privacy via WGAN-GP ensures secure transmission of model parameters without compromising detection performance.