.png){kind=icon}
.png){kind=icon}
Return
Adaptive waiting time asynchronous federated learning in edge computing
.png){kind=icon}
PRE
AI
DOI：10.23919/JCC.fa.2023-0494.202509.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Federated learning combined with edge computing has greatly facilitated transportation in real-time applications such as intelligent traffic systems. However, synchronous federated learning is inefficient in terms of time and convergence speed, making it unsuitable for high real-time requirements. To address these issues, this paper proposes an Adaptive Waiting time Asynchronous Federated Learning (AWTAFL) based on Dueling Double Deep Q-Network (D3QN). The server dynamically adjusts the waiting time using the D3QN algorithm based on the current task progress and energy consumption, aiming to accelerate convergence and save energy. Additionally, this paper presents a new federated learning global aggregation scheme, where the central server performs weighted aggregation based on the freshness and contribution of client parameters. Experimental simulations demonstrate that the proposed algorithm significantly reduces the convergence time while ensuring model quality and effectively reducing energy consumption in asynchronous federated learning. Furthermore, the improved global aggregation update method enhances training stability and reduces oscillations in the global model convergence.
Keywords:
adaptive waiting time
asynchronous federated learning
D3QN
edge computing
Journal
IF:
3.1
Papers:
1.9K
Citations:
5.0K