Verizon this week revealed it is now applying machine learning algorithms against a massive data lake the carrier has built to support Fios fiber optic network services.

Machine learning algorithms are at the heart of applying artificial intelligence (AI) to managing networking. For operators such as Verizon, these tools may be key to reducing the cost of network operations as the amount of data they must handle scales exponentially.

Machine learning is already being used to enable Verizon to identify manufacturing or production defects in third-party hardware and software, says Matt Tegerdine, director of network performance analytics at Verizon.

In addition, Verizon applied machine learning algorithms to daily tests that enabled it to determine that the home router it installs have been consistently providing a one gigabit connection. Previously, Verizon had been marketing that network service as a 750 megabits connection.

Verizon has been leveraging open-source technologies such as Hadoop and the Apache Spark in-memory computing framework to create a data lake for several years.

“We didn’t think it made sense to pay a third-party to mine our own data,” says Tegerdine.

That work was significantly advanced when Verizon acquired Yahoo!, which originally developed Hadoop to analyze the massive amounts of data being generated by the millions of users accessing thousands of cloud services, says Tegerdine. Those collective efforts have significantly reduced the number of false positives that might need to be investigated by network operations teams on any given day, adds Tegerdine. That not only reduces alert fatigue, it allows those teams to concentrate more of their efforts on identifying and resolving more complex issues, adds Tegerdine.

Tegerdine says Verizon is now extending the use of AI-infused algorithms to next predict where fiber lines are most likely to be cut because of, for example, construction in a specific geography. Armed with that data, Verizon can begin to reroute traffic before any customers might be potentially impacted, says Tegerdine.

It may take a while for other providers of networking services to apply AI to network operations. But it’s clear the future of network operations management will increasingly incorporate both machine and deep learning algorithms. That may engender a massive amount of disruption to existing process and cultures within those organizations. But at this point it’s now a matter of when, rather than whether, those changes will be occurring.

---