Research Article
MAML-Based D2D Power Control Scheme in User-Variable Scenario
@INPROCEEDINGS{10.1007/978-3-031-34790-0_28, author={Zhenyu Fan and Xinyu Gu}, title={MAML-Based D2D Power Control Scheme in User-Variable Scenario}, proceedings={Communications and Networking. 17th EAI International Conference, Chinacom 2022, Virtual Event, November 19-20, 2022, Proceedings}, proceedings_a={CHINACOM}, year={2023}, month={6}, keywords={Model-agnostic meta-learning deep neural network user-variable weight initialization optimization}, doi={10.1007/978-3-031-34790-0_28} }- Zhenyu Fan
Xinyu Gu
Year: 2023
MAML-Based D2D Power Control Scheme in User-Variable Scenario
CHINACOM
Springer
DOI: 10.1007/978-3-031-34790-0_28
Abstract
Meta-Learning has been extensively studied since it has the ability of quickly learning new skills by leveraging prior few-shot tasks, which is capable of relieving the problem of relying on large amount of data sample existed in deep learning. In this paper, we apply an algorithm of model-agnostic meta-learning (MAML) to cope with Device-to-Device (D2D) transmit power control issue in user-variable scenario. Specifically, MAML first learns good weight initializations of D2D power control neural network in initial D2D scenario, contributing to a meta-learner. When the number of D2D user changes, the network loads the meta learner and quickly adapts to a new scenario on a few shots of samples. Simulation results demonstrate that MAML shows good performance in generalization and MAML better conducts D2D user-variety power control issues than regular deep neural network power control methods.
