Software-Defined

Machine Learning (ML) and Artificial Intelligence (AI) in 5G networks

by kevin.shatzkamer on ‎02-21-2016 06:00 PM - last edited on ‎02-22-2016 05:10 AM by ShelbyKhan (8,078 Views)

Co-authored by Dave Meyer, Brocade Fellow, and Kevin Shatzkamer, CTO, Mobile Networks

 

Every day there is a new announcement involving big data or real-time analytics for mobile carrier networks. These technologies, along with machine learning, are being advertised as the breakthrough technologies that will enable business and operational transformation for mobile service providers. In fact, several recent technological breakthroughs have, for the first time, enabled the practical machine learning which is rapidly becoming a key component of the mobile carrier’s analytics toolkit.

 

While several recent breakthroughs have made the application of machine learning to the more general field of predictive analytics practical, the standard data analytics of today largely rely on the same tools and technologies that have been in use for the better part of the last century. In contrast, machine learning seeks to use new approaches to discover the processes which generate the datasets we observe. These approaches allows us to powerfully and efficiently classify events of interest and predict some types of network events, such as component failure, congestion, anomaly detection and other security functionality, as well as  other network problems.

 

The same machine learning approaches can be used to understand consumer behavior and how it correlates with events in the network and beyond. These capabilities are key to the carrier’s ability to optimize and operate next generation virtualized SDN and NFV based 5G networks. Machine learning will play a key role these networks in automated detection, prediction and remediation of network resource contention such as congestion or queue utilization, Advanced Persistent Threats and other components of the carrier’s security posture, and emerging technologies such as IoT and M2M.

 

Mobile networks generate vast amounts of the fuel needed for effective machine learning, namely, data.  In addition to traditional data sources, such as NetFlow, and configuration data, such as Chef recipes and Heat templates, the mobile operator’s infrastructure is a rich source of behavioral data. One such source of that behavioral data is the Evolved Packet Core (EPC), which tracks User Equipment (UE) mobility events, idle/active transitions, and traffic flows in order to deliver mobility services to mobile devices. This data can be used to learn the complex statistical models that underlie network, consumer, and device behavior.  

 

The growing machine learning toolbox now includes:

  • Classical statistical models, such as Latent Dirichlet Allocation and Hidden Markov Models
  • Connectionist oriented models, such as Deep Neural Networks

 

Even though many of these technologies have been known for some time, they are just now becoming feasible due important theoretical breakthroughs, the advent of cost effective high performance computing, and abundant data.  Brocade believes that machine learning will, in the very near term, become a key component of every mobile operator’s intelligent infrastructure, and will be foundational to 5G.

 

Industry Proof Points

 

The industry appears to be converging on a similar understanding – that ML/AI will be a foundation technology for mobile networking, especially in 5G. For instance, Cognet, part of the H2020 5G Public Private Partnership (5GPPP) seeking to apply ML algorithms to collected network data to deliver self-organization, management, and fault correction to networking.

 

Brocade continues work on ML/AI initiatives within the industry, as well. For instance, KDDI Laboratories recently announced the World’s First Proof of Concept (PoC) for AI-assisted automated network operations, done in coordination with Brocade, Hewlett-Packard, and Wind River. With the AI-based monitoring system, KDDI Laboratories was able to learn and predict when hardware or software anomalies would lead to catastrophic network failures, and implement appropriate recovery plans through an integrated management system. A distributed AI monitoring capability in the PoC was necessary in order to provide precise analysis on the huge amount of statistical data.

 

While research and development continues this PoC is a live demonstration to the industry of the applicability of ML/AI to mobile network operations. Brocade is proud to be contributing its technology and thought leadership to the industry and working closely with service providers to continue to develop ML/AI capabilities.