.png){kind=icon}
.png){kind=icon}
Return
EECF: An edge-end collaborative framework with optimized lightweight model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129319.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The rapid expansion of the Internet of Things (IoT) has led to an exponential growth in end devices, making edge computing (EC)-based task scenarios more pragmatically significant than cloud computing. However, the inherent limitation in the resources of edge devices impedes the deployment of machine learning models characterized by extensive parameters or intricate structures. To address this conundrum, we contemplate an Edge-end Collaborative Framework (EECF). Specifically, within an image classification task, a CNN model is pre-trained using a subset of representative samples, leveraging the abundant resources at the edge side. Subsequently, the CNN model is deployed to the end side to extract features from the samples, realizing a magnitude reduction in the dimensionality of the samples. Then, an SVM model is deployed at the end side, where a quasi-bird swarm algorithm (QBSA) is utilized for optimizing the parameters of SVM, and the fine-tuned SVM model performs task training based on the dimensionality-reduced samples. Moreover, the analysis of the convergence of QBSA is given to guarantee the successful application. Finally, the performance of EECF is appraised using the Receiver Operating Characteristic (ROC) curve analysis. Empirical results substantiate that the EECF enhances the precision of the model significantly, achieving an uptick in accuracy of up to 10.86 %, compared to benchmarks. In addition, the core code can be accessible via the link: https://tinyurl.com/yuy3snrz .
Keywords:
Edge computing
Internet of Things
CNN
SVM
Quasi-bird swarm algorithm
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W