Machine learning platform architecture demands deep analysis
Article 1 of 4
Model building and usage are both musts in machine learning
Machine learning has become a key enterprise AI building block in companies, Forrester Research analysts Mike Gualtieri and Kjell Carlsson wrote in a September 2018 blog post. And, they said, a variety of platforms are available that can help data scientists "crank out machine learning models" and work with business users to deploy them in production analytics applications.
Those two capabilities are crucial elements of an effective machine learning platform architecture, according to Gartner analysts Carlie Idoine, Peter Krensky, Erick Brethenoux and Alexander Linden.
In Gartner's 2019 Magic Quadrant report on data science and machine learning platforms, they said a 2018 survey conducted by the consulting firm found that more than 60% of finished machine learning models weren't being put to any operational uses. The analysts warned that unless models are embedded in business processes and maintained over time, "the full benefit -- including business value -- of data science and [machine learning] will not be achieved."
One complicating factor for prospective users is that there's a plethora of platform technologies to consider. This handbook provides insight into choosing and using the right machine learning platform.
First, experienced analytics managers discuss the factors behind their choices of machine learning systems and detail the steps they're taking to get useful information from the technologies. We then look separately at two new options for a machine learning platform architecture: incorporating machine learning functionality into IoT equipment and other edge devices as well as running applications on GPU-based systems.