| Title |
Implementation of Device-adaptive Neural Network Generation and Deployment |
| DOI |
https://doi.org/10.5573/ieie.2024.61.1.27 |
| Keywords |
Device-adaptive; Neural network generation; Neural network deployment; MLOps; Neural network model recommendation |
| Abstract |
As device performance improves, artificial intelligence applications are also rapidly being applied. Rapid neural network development suitable for devices determines a company's competitiveness, and to support this, MLOps (Machine Learning Operations), a framework that can be developed and applied immediately, is being provided by global companies that provide cloud services. However, the currently provided framework uses a cloud service that provides high-performance resources for a fee, and there are limitations in developing a neural network optimal for the target device desired by the developer. In this study, a framework was implemented so that optimal neural networks and applications can be developed by considering the devices on which the inference neural network runs. Accordingly, compared to existing frameworks, it includes a function that allows developers to develop neural networks easily and quickly even if they have less expertise in artificial intelligence. Additionally, the developed framework is made public on GitHub, providing source code to interested developers and developers who face difficulties in industrial application. |