微控制器平台低代码神经网络推理技术研究

Research on Low-code Neural Network Inference Technology of Microcontroller Platform

基于微控制器平台的深度学习技术(TinyML)的兴起,意味着在广大嵌入式平台上部署人工智能的重大革新必将带来AIoT的发展。微控制器种类繁多,不同的芯片,其开发方式和基础API也各不相同,与PC端相比,模型的推理技术开发工作难度不小,需要具备一定的嵌入式开发经验。本文尝试将高级语言Python与微控制器结合,为入门开发者提供一套简单、可靠的开源低代码开发套件(名为“OpenART”),可以大大提高在微控制器上部署深度学习的开发效率、降低学习成本。

The rise of deep learning technology on microcontrollers(TinyML)marks a major innovation in the deployment of artificial intelligence on the vast number of embedded platforms,which will greatly boost the development of AIoT.There are many kinds of microcontrollers,and different chip development methods and basic APIs are also different.Compared with the PC side,the development of the reasoning technology of the model is not easy,and certain embedded development experience is required.We try to combine high-level language Python with microcontrollers to provide beginner developers with a simple and reliable low-code development kit called"Open-ART",which can greatly improve the development efficiency and reduce learning efforts of deploying,evaluating,and benchmarking deep learning on microcontrollers.