TinyML

Chapter 18. Debugging

You’re bound to run into some confusing errors as you integrate machine learning into your product, embedded or otherwise, and probably sooner rather than later. In this chapter, we discuss some approaches to understanding what’s happening when things go wrong.

Accuracy Loss Between Training and Deployment

There are a lot of ways for problems to creep in when you take a machine learning model out of an authoring environment like TensorFlow and deploy it into an application. Even after you’re able to get a model building and running without reporting any errors, you might still not be getting the results you expect in terms of accuracy. This can be very frustrating because the neural network inference step can seem like a black box, with no visibility into what’s happening internally or what’s causing any problems.

Preprocessing Differences

An area that doesn’t get very much attention in machine learning research is how training samples are converted into a form that a neural network can operate on. If you’re trying to do object classification on images, those images must be converted into tensors, which are multidimensional arrays of numbers. You might think that would be straightforward, because images are already stored as 2D arrays, usually with three channels for red, green, and blue values. Even in this case, though, you do still need to make some changes. Classification models expect their inputs to be a particular width and height, for example 224 ...