AI at the Edge

Chapter 10. Evaluating, Deploying, and Supporting Edge AI Applications

This is the final theory chapter of this book, and we’ll be covering the processes of evaluating, deploying, and supporting edge AI applications. These three things are intimately connected—and in an iterative project they happen in parallel throughout the course of the development workflow.

Evaluating Edge AI Systems

Evaluation is the key to a successful project. In fact, without evaluation you have no real way of knowing whether your project is successful or not. While it happens to be featured at the end of this book, it’s something that occurs all the way through the development process. It may even start before development, when you’re quantifying the performance of an existing process you aim to improve with edge AI.

Remember that throughout the process, evaluation needs to be conducted with participation from stakeholders and end users. It’s very possible that different stakeholders may have conflicting criteria for evaluation, depending on their individual perspectives. Figuring out how to resolve these conflicts is an important part of evaluation.

Here are some of the stages where evaluation needs to occur:

Examining an existing solution

Much of the time, we’re developing edge AI systems to replace legacy systems that we think could be better. This makes it very important to understand the actual performance of the existing system at the beginning of the process. The goal of our development process ...