ML Map v0.1 - Roles of Machine Learning

Machine Learning is a huge subject space, it's useful to have a map to navigate it, and find your place.  In this article, I’m going to attempt to define this map in what will probably be the first of many iterations. So welcome to v0.1 of the ML Role Map.  In particular, I want to focus on MLOps. 

For this map, I'm going to create a spectrum. A line of three distinct roles within ML and how they relate to each other. The spectrum is useful as it removes hard borders between roles.  However, it's important not to read any hierarchy or status into these roles.

From left to right:

Developer

This is not, in this case, specifically an ML role, but the field of ML needs this role to integrate ML workflows and services into end-user applications.   For example, integrating a recommendation engine into a mobile app. The Developer needs to work in conjunction with the ML Engineer to facilitate this.

ML Engineer

The roles main responsibility is to derive real business (or social) value from ML. Of all the roles I define here, this is the one that will be further broken down in subsequent versions of the map. As such there could be a great number of examples I could highlight but for now let's imagine this ML Engineer processing data, training a model and deploying it to a production endpoint. (Yes, that's many things, like I say we will break this down at a later date.)

ML Researcher

This role is the longest-lived of all the ML roles. These roles have been around in effect since the 1960s and only with the advent of affordable computing power has the contemporary field of ML formed 'around' this role.  Today the ML Researcher may well be based in academia and will spend their time advancing our understanding of ML through statistical analysis and the design and refinement of ML algorithms and processes.

MLOps

Around the centre of the map is MLOps. And while I don't wish to imply any hierarchy or status to the roles themselves, I will argue that MLOps is at the centre of all of ML, and is very much a driving force.

MLOps (or Machine Learning Operations) is a term derived from DevOps (Developer Operations) and is somewhat a result of our ability to deploy infrastructure as code, mostly thanks to cloud platforms such as AWS.

MLOps, therefore, is the operationalizing of ML from research into the hands of end-users (whether that be in a mobile app or a medical diagnosis). 

Simply put, MLOps is key to realizing the benefits of ML in the real world.

As such, I will spend most of my time focusing on MLOps, and issues around MLOps in subsequent articles I write, the commentary I make, and the courses I produce.

Finally

In case you don't know, I'm making a course specifically focused on helping students pass the AWS Machine Learning Specialty Certification (MLS-C01). And while it's not in the name, this exam is squarely focused around MLOps.

I imagine this post will raise as many questions as it answers. So please get in touch through comments or directly to discuss further. Please let me know if you like this post, and also share it with anyone you think might be interested.