Last week I posted Machine Learning Platforms for Predictive Applications: Part 1 – Introduction. That post provided an introduction to a series of articles, the second of which you’re reading right now, on the topic of Machine Learning Platforms. If you haven’t yet read the introduction, go ahead and take a look; we’ll be here when you get back.
In this post I answer your burning question, “so, what is a Machine Learning Platform anyway?”
I define a Machine Learning Platform as:
“A platform for automating and accelerating the delivery lifecycle of predictive applications capable of processing big data using machine learning or related techniques.”
While this definition probably strikes you as straightforward enough, there’s quite a bit there when you begin to peel back the layers. A few of the key ideas captured in this definition are:
- Acceleration. Acceleration has two meanings for Machine Learning Platforms: (a) Acceleration of the solution delivery lifecycle (the time it takes to deliver new predictive applications), and (b) run-time acceleration achieved through technical advances such as distributed and in-memory computing.
- Automation. The real-world job of the data scientist or analyst consists of many repetitive and time-consuming tasks. Automating these tasks can eliminate project bottlenecks, allowing organizations to deliver new projects more quickly, update existing projects more frequently, and take on more projects without increasing staffing.
- Big Data. While I won’t venture to define big data here, suffice it to say that in order to be truly useful, a Machine Learning Platform must be capable of helping users process large volumes of data from a wide variety of sources.
- Full-Lifecycle. Machine learning is not new, and there are many technologies that help enable it. Machine Learning Platforms differ from desktop tools and software libraries in that their focus is on enabling the full lifecycle of delivering predictive applications. More on this to come.
- Platform. Machine Learning Platforms are geared towards enabling other software applications to make better decisions or recommendations, and thus must easily integrate with them.
- Predictive. Machine learning can be applied to different types of problems, but Machine Learning Platforms focus on helping businesses predict future outcomes such as a customer’s propensity to buy when presented with a given offer, or the likelihood that a given transaction will be rejected as fraudulent.
Stay tuned for next time when I talk about what Machine Learning Platforms are NOT.
To make sure you don’t miss anything, please subscribe to receive updates.