IBM will add deep learning for mainframes

Mainframes operate in the largest companies in the world, including banks, insurance companies, retailers and airlines. Despite the increasingly popular cloud services, mainframes remain in service due to their tremendous performance. For example, the modern IBM z Systems mainframe is capable of processing about 2.5 billion transactions per day in real time - this is the equivalent of transactions that would be processed for 100 cyber-monitors.

Now IBM plans to add support for part of its cognitive services for mainframes for more efficient data processing, offering in-depth training for all users of its mainframes. It is planned to add this feature to any technology company that is related to big data when information is stored behind a firewall, which IBM calls the "private cloud."

“About 90% of the information in our world cannot be obtained using search services like Google. This data is hidden behind firewalls in private clouds. How can we automate the processing of such data sources? ”Asks IBM Research Manager Rob Thomas.
')
IBM plans to provide data workers with the same mainframe capabilities they get when analyzing BigData using cloud services. The main goal of such a mission is to automate the routine work of creating, testing and deploying analytical models of various types. IBM solutions are compatible with popular open-source projects like Apache Spark ML, Tensor Flow and H2O. The important point is that mainframes will be able to work with any types of data that the client has.

Seasoning for all products is IBM Research's Cognitive Assist for DataScience. This product helps to choose the optimal data processing algorithm by analyzing the entire list of available algorithms. For each of the data types, the most efficient algorithm will be chosen that will allow you to quickly get the desired result.

Over time, the system will become more “smarter” thanks to the possibility of self-learning. The computer will be able to “understand” the degree of efficiency of each of the algorithms with respect to a specific data set. “This will allow data engineers to create the desired model, and the IBM Machine Learning technology will determine the best algorithm. The more data will be processed, the smarter the algorithm will become, ”said Rob Thomas.

Experts have developed prototypes of a weak form of artificial intelligence and machine learning technology for mainframes a long time ago, several decades ago. But technology does not stand still, and the product offered by IBM for mainframe is now tens of thousands of times more efficient than the first platforms of this type. In addition, modern solutions are less expensive, because, firstly, they are largely based on open source software, and secondly, they can work in a fully automatic mode, almost without the participation of people.



New features will be available to mainframe users already this quarter. As mentioned above, gradually the weak form of AI will be added to almost all IBM software products. Among other things, new features will be gained by the IBM z Systems mainframe family. Here the most powerful system is the z13 . That it is able to handle about 2.5 billion transactions per day, as mentioned above. The technologies used in the z13 allow you to detect fraud cases in real time for 100% of business transactions, instantly providing the user with analytical data to evaluate the operation.