Autonomous Network Module Orchestration for Coupled AI/ML Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 6G network technologies face challenges in training and inferring strongly coupled AI/ML models for autonomous network functions, leading to elevated sampling and computational costs, as well as substantial memory and storage requirements due to interdependent functions sharing common input metrics and output control parameters.
Innovation Solution
A modular, programmable, and standards-compliant network automation system that organizes AN functions as interchangeable modules, enabling efficient building, orchestration, and adaptation of interdependent functions through module-based graphs and automated routing, reducing computational and storage costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If strongly coupled AI/ML models are trained and inferred for autonomous network functions, then network automation capabilities are enhanced, but sampling and computational costs increase
Solution Approach 1:
The patent segments autonomous network functions into modular, interchangeable components that can be independently trained and inferred. This modular architecture allows the system to selectively activate only the specific modules needed for particular network tasks, avoiding the computational overhead of running all AI/ML models simultaneously. The segmentation enables efficient resource utilization while maintaining network automation capabilities.
Solution Approach 2:
The patent implements a universal module repository that stores reusable AI/ML modules capable of serving multiple network functions. These modular components can be dynamically selected and composed to address different autonomous network requirements, eliminating the need to train separate specialized models for each function. This multi-functionality reduces overall computational costs while enhancing automation across diverse network scenarios.
2Extent of automation
If strongly coupled AI/ML models are trained for autonomous network functions, then network automation is improved, but memory and storage costs increase
Solution Approach 1:
The patent divides autonomous network functions into discrete, stored modules in a centralized repository. Each module represents a reusable computational unit that can be retrieved and executed as needed. This segmentation allows the system to store only essential modular components rather than maintaining complete models for all possible network functions, thereby reducing memory and storage requirements while preserving automation capabilities.
Solution Approach 2:
The patent implements a module repository system where standardized AI/ML modules are created once and then copied or referenced multiple times across different network functions. Instead of storing redundant full models for each function, the system stores single instances of modular components that can be instantiated and executed repeatedly, significantly reducing storage costs while maintaining full automation functionality.
3Reliability
If independent training and inference of diverse AI/ML models are performed for specific functions, then function-specific optimization is achieved, but training and computational expenses increase
Solution Approach 1:
The patent segments AI/ML training into modular units that can be independently developed and stored in a repository. Each module can be trained separately for specific functions, but the modular structure allows these trained components to be reused across multiple applications. This segmentation enables function-specific optimization while reducing total training time, as modules need to be trained only once rather than repeatedly for each function.
Solution Approach 2:
The patent implements preliminary training of AI/ML modules and stores them in advance in a centralized repository. When autonomous network functions need to be executed, the system retrieves pre-trained modules rather than training models from scratch. This preliminary action significantly reduces training time and computational expenses while maintaining function-specific optimization, as the heavy training work is performed beforehand and reused as needed.
4Adaptability or versatility
If modular, programmable network automation system is implemented, then adaptability and coordination among AN functions are improved, but system complexity increases
Solution Approach 1:
The patent segments the network automation system into standardized, interchangeable modules with well-defined interfaces. This segmentation creates a manageable architecture where complex autonomous functions are broken down into smaller, independent components. The modular structure improves adaptability by allowing flexible composition of modules while reducing perceived system complexity through clear separation of concerns and standardized interaction protocols.
Solution Approach 2:
The patent implements a universal module repository and standardized interface framework that enables the same modular components to serve multiple autonomous network functions. This universality reduces system architecture complexity by providing a common platform and interface standard, eliminating the need for separate specialized architectures for each function. The standardized framework simplifies module integration and interchangeability while enhancing overall system adaptability.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Disclosed are devices, methods, apparatuses, and computer-readable media for autonomous networks. An example apparatus for an autonomous network service provider may include at least one processor and at least one memory. The at least one memory may store instructions that, when executed by the at least one processor, may cause the apparatus at least to: receive a first request for a first application; retrieve by metadata retrieval function, first metadata for the first application; and retrieve by module retrieval function, one or more first modules of input, one or more first modules of function and at least one first module of task according to the retrieved first metadata.