CSPs must develop ‘green AI’ capabilities to process increasingly complex datasets, support autonomous network operations and reduce the industry’s carbon footprint.
Green AI for autonomous networks
As CSPs continue their digital transformation, they become capable of providing increasingly complex and precise services – however these in turn demand more complex modelling, as well as greater data annotation, computing power, training time, and ultimately resources. This set of demands applies to various business requirements, each requiring their own AI modelling – yet modelling for each instance is a barrier to joining and systematizing the knowledge and experiences of each project. What’s needed is a ‘green’ AI that can make use of as much data as possible, making the computing of autonomous networks more efficient and ultimately helping organizations reduce their carbon footprint.
This is the purpose the Green AI for autonomous networks Catalyst, a collaborative project involving China Telecom, China Unicom, Inspur, Keda Guochuang, and AsiaInfo. The principal focus is how vertical domain knowledge graphs can be shared and applied between network topologies to build autonomous knowledge graphs which improve the network’s perception, analysis, decision-making and execution processes. By using data such as that from base station KPIs, alarms, cases, work orders and business information and expert knowledge, an enhanced network knowledge map is constructed, so deep learning and big data analysis can be used for continuous, iterative optimizations.
Using knowledge graphs to reduce energy consumption
A key aspect of the Catalyst involves developing a method to extract knowledge and relationships from semi-structured and unstructured data that can be used to build knowledge graphs appropriate to the autonomous network. This knowledge is then sublimated into ‘wisdom’ which is used to realize knowledge applications such as retrieval, recommendation and interaction and ultimately support autonomous perception, planning, optimization and healing.
In so doing, the solution realizes energy saving in three key ways:
Using knowledge graphs to improve autonomous operations
The network knowledge graph can provide prior knowledge for AI modelling, reduce data labelling, and provide prior knowledge features to assist model training, helping to save resources across business. These efficiencies can also achieve scale the more they are used. For example, after being applied to one business application, some resulting knowledge can be reused to meet new requirements. At the same time, experts and analytics can feed back their insights into a systematic network knowledge graph system. Supporting this process is the enhanced knowledge retrieval where the knowledge graph replaces manual document retrieval, manual analysis and decision-making, which helps the CSP to correlate and diagnose issues and formulate the best solution much more efficiently.
Autonomous networks are a key part of how CSPs’ digital transformation – but a crucial component of their realization is the advanced and energy-efficient computing which we now call green AI. At a time when the conversation of resources is as important as ever, green AI helps provide a route to smart, sustainable growth.