IBM: Machine Learning and the Mainframe

The drumbeat of services, press releases and strategic partnerships related to machine learning processes and deep learning frameworks shouldn’t surprise anyone with an eye on IT markets. Both machine and deep learning efforts reflect the increasing interest and momentum around business use cases for augmented and artificial intelligence.

That said, less attention has been given to some practical processes underlying machine learning that are crucial to its commercial success. One example; like virtually all analytically-intensive processes, the volume of information machine learning processes utilize directly impacts their overall value.

In other words, the more data you can access, the clearer your results can be. And when I say “more,” I’m not talking about mere hundreds of gigabytes of data, but tens and hundreds of terabytes or even petabytes of validated information.

Another example: even experienced data scientists can have a hard time choosing the ideal algorithm for a machine learning project. Plus, project managers may disagree about the best language or framework to use, resulting in potentially lengthy and costly delays.

These issues can impact a project’s viability, but so do common, practical processes. Moving enormous data sets into dedicated analytics systems takes time and effort, can impact security and may be undermined by inadvertent human errors. Similarly, data scientists can waste days or weeks in developing, testing and retooling an analytic model if better ones exist.

These points cast light on the new IBM Machine Learning offering and the company’s decision to deliver it initially for the zSystem mainframe platform. So it’s worth considering the likely enthusiasm among enterprise customers and data scientists for IBM’s new solution.

Machine learning and the mainframe

Why so? Primarily because IBM Machine Learning (which was extracted out of the Watson cognitive platform) is designed to seamlessly utilize the vast stores of information that reside in zSystem infrastructures.

How much data are we talking about? Some estimate that as much as 80 percent of the world’s business information is stored in mainframe systems. As a result, the banks, government agencies, insurers, retailers and transportation companies that currently use IBM’s zSystem to perform billions of transactions daily can jump into machine learning projects quickly and without significant additional investments.

Those companies can also utilize IBM Machine Learning to train models on historical and current business information, then perform real time transaction scoring, significantly increasing the applicability of results. That makes IBM Machine Learning useful for numerous dynamic business scenarios, including retail sales forecasting, complex supply chain management, financial analysis and recommendations, and personally tailored healthcare services.

The company has incorporated other features that heighten IBM Machine Learning’s usability and value, including initially supporting a popular language called Scale on Spark (other languages and frameworks will be added eventually), and any transactional data type. The new platform will also leverage Cognitive Automation for Data Scientists, a new solution from IBM Research that scores data and project requirements against available algorithms, then recommends the best option.

IBM Machine Learning will be commercially available in March as an on-premises private cloud offering on zSystem infrastructures. The company plans to add support for the platform on its Power Systems next, then will deliver a related Data Science Experience solution on IBM Cloud supporting hybrid cloud implementations.

Final analysis

What’s the takeaway from all this? There are three points to consider. First, the new Machine Learning platform demonstrates how IBM captures unique synergies through its strategic investments. In this case, that includes the company’s internal organic R&D efforts and external acquisitions in advanced analytics, as well as evolutionary zSystem developments.

In addition, IBM Machine Learning is poised to offer significant benefits to and create new business opportunities for both the company and its clients. Finally, IBM’s Machine Learning underscores the enterprise value of its zSystem mainframe and Power Systems platform. Since most other IT vendors have abandoned homegrown innovations for off the shelf industry standard components, that is an important competitive differentiator.

Overall, IBM Machine Learning should prove valuable for many of the company’s enterprise customers. Considering how rapidly interest is growing for leveraging machine learning in business scenarios, the new IBM platform’s affinity for zSystem environments and eventual support for Power Systems and IBM Cloud should inspire strong interest among the company’s core clients.