TinyML

Chapter 14. Designing Your Own TinyML Applications

So far, we’ve explored existing reference applications for important areas like audio, image, and gesture recognition. If your problem is similar to one of the examples, you should be able to adapt the training and deployment process—but what if it isn’t obvious how to modify one of our examples to fit? In this and the following chapters, we cover the process of building an embedded machine learning solution for a problem for which you don’t have an easy starting point. Your experience with the examples will serve as a good foundation for creating your own systems, but you also need to learn more about designing, training, and deploying new models. Because the constraints of our platforms are so tight, we also spend a lot of time discussing how you can make the right optimizations to fit within your storage and computational budgets without missing your accuracy targets. You’ll undoubtedly spend a lot of your time trying to understand why things aren’t working, so we cover a variety of debugging techniques. Finally, we explore how you can build in safeguards for your users’ privacy and security.

The Design Process

Training models can take days or weeks, and bringing up a new embedded hardware platform can also be very time-consuming—so one of the biggest risks to any embedded machine learning project is running out of time before you have something working. The most effective way to reduce this risk is by answering as many of ...