Practicing Trustworthy Machine Learning

Chapter 6. More State-of-the-Art Research Questions

Much of what we’ve covered in previous chapters has ranged from standard practices to practical yet underutilized methods. This chapter is dedicated to methods that are just exiting the research phase. We will also go into which of these methods are becoming practical for the real world and ask how different they are from the various trustworthiness metrics we’ve discussed in the previous chapters. This is by no means an exhaustive list of some of the various bleeding-edge ML techniques in the works right now. However, these are some of the more interesting techniques the authors have seen come up in discussions.

First, we want to go over how to watch out for overhyped reports and articles about machine learning techniques (which will be extremely relevant in “Quantum Machine Learning”).

Making Sense of Improperly Overhyped Research Claims

In general, all of these techniques are still in their research and proof-of-concept stages. It would help if we had more examples of how to judge research techniques and their readiness for the real world. This might sound like an oxymoron, since the whole point of machine learning research is to produce insights and techniques beyond what was previously possible.

A more helpful approach might be to look for red flags in reports on new machine learning advances. Indeed, given how fast the field of machine learning has been moving, it can be difficult to enforce a standard or culture of quality ...