Press Releases

Machine learning the solution to networking problems

September 7, 2018

Posted by: Anasia D'mello

Computers can now learn to solve networking problems for themselves, a study from the University of Waterloo has found.

“There are several challenges in computer and telecommunication networks, such as network security, network management, and traffic engineering,” said Raouf Boutaba, a professor in the David R. Cheriton School of computer science at Waterloo. “The use of machine learning in this context will make networks more secure, better managed, and provide a better quality of service.”

Today’s networks, whether wired or wireless, are increasingly complex and diverse. Network managers ensure that devices remain connected, applications run, the volume of traffic flows smoothly, data is collected, and that the system remains secure. Disruptions to the network can be costly and time-consuming.

In conducting the study, an extensive literature review, the researchers found machine learning can address current human-centric network activities, which tend to be costly, error-prone and slow to adapt to changes. Those challenges, they found, could be overcome by giving network administrators the ability to automate network management, as well as the ability to forecast how networks will behave under specific circumstances.

“Machine learning would work hand-in-hand with humans, where humans set the policies and the machine learning-enabled network controller derive the proper executable actions to monitor, operate and manage the network,” said Boutaba. “Humans are also required to establish accurate machine learning models to be used for prediction and inference.”

A comprehensive survey on machine learning for networking: evolution, applications and research opportunities appears in the Journal of Internet Services and Applications. Led by Boutaba, the survey was compiled by Cheriton School of Computer Science researchers Mohammad A. Saluahuddin, Noura Limam, Sara Ayoubi, Nashid Shahriar, and co-authored by University of Cauca researchers Felipe Estrada-Solano and Oscar M. Caicedo.

Comment on this article below or via Twitter @IoTGN