Edge-computing-assisted intelligent processing of AI-generated image content
Suzhen Wang,Yongchen Deng,Lisha Hu,Ning Cao,Cao, Ning
DOI: https://doi.org/10.1007/s11554-023-01400-w
IF: 2.293
2024-02-25
Journal of Real-Time Image Processing
Abstract:Artificial intelligence-generated image content (AIGIC) is produced through the extraction of features and patterns from a vast image dataset, requiring substantial computational resources for training. Due to the limited computational resources of terminal devices, efficiently processing and responding to AIGIC has emerged as a critical concern in current research. To address this challenge, the present paper proposed the utilization of edge computing technology. Edge computing enables the offloading of certain training tasks to edge nodes, facilitating expedited task offloading strategies that empower terminal devices to generate image content efficiently. Building upon the edge serverless architecture, this paper introduces an edge serverless computing framework based on Web Assembly (WASM). Notably, this framework effectively resolves the latency issue associated with container cold start in serverless computing. Additionally, to enhance the collaborative capabilities of edge nodes, entropy-based Proximal Policy Optimization (E-PPO2) is proposed. This algorithm enables each edge node to share global rewards, continually update parameters, and ultimately derive the optimal response strategy, thereby harnessing edge device resources to their fullest potential. Finally, the efficacy of the proposed serverless computing architecture based on WASM is demonstrated through the evaluation of 13 benchmark functions. Comparative analyses with four task offloading algorithms highlight that the E-PPO2 algorithm, proposed in this article, significantly reduces task execution latency, facilitating rapid processing and response in AIGIC scenarios.
computer science, artificial intelligence,engineering, electrical & electronic,imaging science & photographic technology