AI at the Edge

Chapter 6. Understanding and Framing Problems

The next five chapters provide a roadmap for working with edge AI. We’ll establish best practices for:

• Viewing the problems you want to solve through the lens of edge AI

• Building datasets that allow you to you train models and evaluate algorithms

• Designing applications that make use of edge AI technologies

• Developing effective applications through an iterative process

• Testing edge AI applications, deploying them, and monitoring them in the field

For this chapter in particular, we’ll start by introducing a high-level, general workflow for edge AI projects. This should give you a sense of how everything will fit together. After that we’ll learn how to evaluate projects to make sure they are a good fit for edge AI, then walk through the process of identifying which types of algorithms and hardware make sense for a given problem—and start to think about planning our implementation.

The Edge AI Workflow

Like any sophisticated engineering project, a typical edge AI project involves multiple tracks of work, some of which run in parallel. Figure 6-1 shows them in context.

Figure 6-1. The edge AI workflow, grouped into the “discover” and “test and iterate” stages

The process can be split roughly into two chunks—labeled in the diagram as discover and test and iterate. The first chunk, discover, involves developing a deep understanding ...