Cognitive network for adaptive intent-based operations

Read the article

Cognitive network

The cognitive network operates with full autonomy capable of continuous self-learning at scale. By taking advantage of machine learning and machine reasoning, it will use the knowledge it has gathered in the past to solve new types of problems without involving humans in the loop.

The vision of cognitive network

There are two main aspects of the cognitive network.

Zero-touch deployment and operation: In our vision of the future networks, we will have many different services, which are deployed and adapted dynamically. Zero-touch means that once a service is defined, all the stages of its life cycle will be managed automatically.
Continuous real time performance improvements: Some services have extreme performance requirements and need to be carefully tuned to meet them. We also need to make sure that the overall network performance is optimized and improved based on performance measurements.

With these requirements, it is no longer possible to configure services and infrastructures manually, where humans make all the detailed configurations and fine-tuning. Instead, the network needs cognitive capabilities: Learning from data, making intelligent choices, and applying these choices in the infrastructure. This applies to all parts of the system, from the antenna site and radio-near configurations to end-to-end service parameters.

The human operator will now interact with the network using high-level requirements in the form of intents. The network then needs intelligence (the ”brains”) to understand these intents and translate them into action plans and settings to fulfill all requirements that are given by the human. To take decisions based on the current knowledge, improve the decision-making process, and acquire new knowledge, data from operations is needed. Finally, all actions and configurations are applied in the infrastructure, which is evolving into a software-based system deployed in a distributed cloud-native platform.

Data-driven operations

A data-driven network architecture is an architecture where relevant data can efficiently and securely be distributed and analyzed with the purpose to optimize network performance and network operations. Data-driven simply means that decisions are made on data (to a larger extent than before). A data-driven architecture is an inevitable infrastructure to enable ML/AI and eventually also MR/AI.

Distributed intelligence

Telecommunications networks are geographically distributed by nature. This implies that, when using novel AI techniques, they need to have native support for distribution.

Continuous learning

A fully automated system will require many closed loops that control different parts of the network infrastructure and the services provided to customers. We need solutions to deploy and optimize these closed loops in a running network. Learnings from operations and service performance are fed back in short cycles or near real time to improve configurations, processes, and software.

Intent-based management

High-level declarative languages simplify many aspects of network operations. Today, there are many domain-specific languages (for example, Kubernetes HELM, TOSCA). Beyond these, we will use intents to describe the goals for system operation. We need corresponding tools to translate the high-level intents into lower-level instructions or direct system configurations.

Explainable and trustworthy AI

Explainability means that the automated system can explain (to a human) why a certain action is taken. Explainability is one aspect that encourages human beings to perceive a system as trustworthy. Trustworthy AI should be: Lawful - respecting all applicable laws and regulations; Ethical - respecting ethical principles and values, and Robust - both from a technical perspective and by taking into account its social environment. When a complex system has a predictable behavior, the trust in the system increases.