A Federated Learning Latency Minimization Method for UAV Swarms Aided by Communication Compression and Energy...

Area of Science:

• Robotics and Autonomous Systems
• Machine Learning
• Network Computing

Background:

• Unmanned aerial vehicle swarms (UAVS) leverage machine learning (ML) for tasks like detection and mapping.
• Federated learning (FL) is suitable for UAVS due to connectivity challenges with ground stations.
• Latency-sensitive operations, such as emergency obstacle avoidance, pose challenges for FL in UAVS.

Purpose of the Study:

• To analyze energy consumption and latency sensitivity of FL in UAVS.
• To propose solutions for optimizing FL in latency-critical UAV applications.
• To enhance edge network computing for UAV swarms.

Main Methods:

• Developed an efficient asynchronous federated learning mechanism for edge network computing (EAFLM).
• Integrated ant colony optimization (ACO) with EAFLM to manage UAV participation and resource allocation.
• Screened UAVs for communication rounds and adjusted transmit power and CPU frequency for faster iterations.

Main Results:

• Significantly reduced communication times between UAVS.
• Demonstrated a relatively low impact on model accuracy.
• Optimized the allocation of UAV communication resources effectively.
• Validated the proposed method using the MNIST dataset.

Conclusions:

• The proposed EAFLM combined with ACO offers an effective solution for latency-sensitive FL tasks in UAVS.
• This approach balances communication efficiency, energy consumption, and computational performance.
• It enhances the practical applicability of FL in dynamic UAV swarm operations.