Core Network Function Orchestration With Federated Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Centralized learning in network data analytics functions (NWDAFs) in 5G core networks faces issues of latency, security vulnerabilities, high implementation cost, and lack of robustness due to data exposure and processing on a central server.
Innovation Solution
Implementing a multi-agent federated reinforcement learning approach that enables distributed machine learning model training between a server and local network functions, using encryption and weighted federated averaging to secure and optimize data transmission and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If centralized learning is applied in the core network with a central NWDAF server, then large amounts of data can be processed centrally, but latency increases significantly due to data transmission and centralized processing time
Solution Approach 1:
The patent segments the centralized learning system into multiple distributed NWDAF agents deployed across different network locations. Each agent independently processes local data, eliminating the need to transmit all data to a central server. This segmentation maintains data processing capacity while reducing latency by enabling parallel processing across multiple nodes.
Solution Approach 2:
The patent transitions from a single-dimensional centralized processing architecture to a multi-dimensional distributed architecture. Multiple NWDAF agents operate across different spatial dimensions (network locations) and temporal dimensions (synchronized iterations), allowing simultaneous data processing that reduces overall latency while maintaining processing capacity.
2Quantity of substance
If local datasets are sent to the central NWDAF server for processing, then centralized learning can be performed, but security and privacy are compromised due to data exposure
Solution Approach 1:
The patent extracts the sensitive local datasets from the transmission process and keeps them securely stored at their source locations. Only the trained machine learning models (which contain no sensitive data) are transmitted between agents and the orchestrator. This extraction maintains data accessibility for learning purposes while eliminating security and privacy risks associated with data exposure.
Solution Approach 2:
The patent introduces an intermediary mechanism where local datasets are processed in-place by NWDAF agents, and only the resulting models (intermediary artifacts) are shared. This intermediary approach allows centralized learning objectives to be achieved without direct exposure of sensitive data, thereby maintaining both data accessibility and security.
3Quantity of substance
If all network data is transmitted to and processed on the central server, then comprehensive analytics can be generated, but implementation cost and energy consumption increase extremely
Solution Approach 1:
The patent segments the data processing workload across multiple distributed NWDAF agents rather than concentrating it on a single central server. Each agent processes only local data, dramatically reducing the data volume that needs transmission and processing at any one location. This segmentation maintains comprehensive analytics capability while reducing energy consumption associated with data transmission and centralized processing.
Solution Approach 2:
The patent enables each local NWDAF agent to perform self-service processing of local data without requiring transmission to a central server. Each agent independently trains models using local datasets, eliminating the energy-consuming data transmission process while maintaining the ability to generate comprehensive analytics through model aggregation.
4Quantity of substance
If a centralized framework processes all data on the server, then a unified model can be trained, but robustness decreases due to lack of control in the learning process and vulnerability to attacks
Solution Approach 1:
The patent segments the learning process into distributed agents that independently process data locally, reducing the attack surface for any single point of failure. Each agent maintains control over its local data and processing, enhancing robustness while the orchestrator coordinates to achieve unified model training through model aggregation rather than data centralization.
Solution Approach 2:
The patent implements a feedback mechanism where the orchestrator receives trained models from local agents, aggregates them, and provides feedback in the form of updated global models back to the agents. This feedback loop maintains control over the learning process and ensures robustness by enabling continuous monitoring and adjustment without requiring centralized data processing.
Data Source
AI summary
The discussed solution provides a framework for managing distributed network functions in a core network. The frame-work may use multi-agent federated reinforcement learning to orchestrate the distributed network functions in the core network. The framework may comprise a server network function responsible for providing a global machine learning model to one or more local network functions and managing the local network functions using a feedback mechanism. The local network functions may perform local training and apply the feedback from the server network function.