Learning efficiency maximization in UAV-and-RIS-aided mobile edge learning system

With the ever-increasing number of Internet of Things devices (IoTDs) and the rapid development of artificial intelligence (AI) technologies, mobile edge learning (MEL) has emerged as a new communication network paradigm that can deploy machine learning on mobile edge computing (MEC) platforms with abundant computational resources. Then how to fully exploit the potential of MEL and enhance its performance is an important issue. To address this issue, the paper proposes an unmanned aerial vehicle (UAV)-and-reconfigurable intelligent surface (RIS)-aided MEL system. In the system, the UAV equipped with a MEL server can fly close to the IoTDs to collect their data for training a deep learning model. And the RIS mounted on the building can improve the wireless channel environment between the UAV and IoTDs to assist the UAV in collecting data. In order to maximize the MEL learning performance while minimizing the system energy consumption, this paper also proposes a new optimization metric called learning efficiency. Then, a learning efficiency maximization problem based on the proposed system is formulated by jointly optimizing the minority class sample size, the transmit resource of the IoTDs, the phase shift of the RIS, and the trajectory of the UAV. Considering the intractability of the problem, we solve it using the alternating optimization (AO) algorithms based on the two types of UAV trajectory design, i.e., a time-division-multiple-access (TDMA) design with higher performance and a Flight-Hover design with lower complexity. The simulation results demonstrate that the proposed optimization metric and algorithms are effective and perform excellently compared to other baselines.