Edge AI paves the way to many new application possibilities across all markets. Yet, for edge AI solutions to reach their full potential, developers need user-friendly solutions that provide accurate performance insights to overcome the complexity of working with machine learning algorithms.

Today, many edge AI applications running on MCUs fail to reach production stage and commercial deployment because existing edge AI solutions lack flexibility and reliability. Overall, they do not allow embedded developers to create, test, and deploy machine learning algorithms on MCUs in an easy and timely way.

One-hour webinar: everything you need to know to integrate AI into your MCUs

In this one-hour webinar experts from ST and NVIDIA will show you how to deploy high performance edge AI algorithms on STM32 MCUs seamlessly.

We will start from a complex, memory-intensive neural network, which is not suitable for embedded AI. By using the NVIDIA TAO Toolkit, we will reduce its memory footprint by a factor of 100 so it can run on an STM32 MCU.

From theory to practice, edge AI experts will also present you a demo on person presence detection.

After attending this session, you will know how to create from scratch an end-to-end edge AI application using NVIDIA TAO Toolkit and how to get it up and running on an STM32 in just one hour. You will also know how to create your edge AI application using your own ANN models, datasets, and embedded software application code.

With this webinar session, you will learn:

• About ST’s ecosystem for deploying edge AI on MCUs
  
• About the NVIDIA TAO tool that lets developers train, adapt, and optimize neural network models
  
• How ST and NVIDIA tools can help you boost edge AI performance on STM32 MCUs in a simple way, outperforming other solutions available on the market
  
• How these tools disrupt the current practices and make edge AI accessible to all
  
• How to get started easily with the ST and NVIDIA AI ecosystem
  

Who should attend?

This session is designed for engineers who:

• already have prior knowledge of AI and neural networks
  
• have heard about the main data science frameworks and file formats (TensorFlow, Keras, ONNX, …)
• have been working with datasets to train neural network models
• are aware of the challenges and limitations to deploy ML/AI on MCUs and are looking for solutions to simplify your project development

Agenda

• Webinar objectives, industry status and ST edge AI solutions
  
• NVIDIA presentation of the TAO Toolkit
• ST hands-on demo of the NVIDIA TAO Jupyter Notebook showing how to:
• deploy a person detection use case starting from with a BYO model
  
• training it with an NVIDIA dataset
• pruning and retraining the network to reduce its memory footprint by a factor of 100 for optimized performance on STM32
• deploying the trained NN using ST edge AI solutions