NWDAF ML Provisioning With Consumer Control in Federated Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current solutions for federated learning among multiple Network Data Analytics Functions (NWDAF) instances in wireless communication networks lack consumer control over the model provisioning and training processes, limiting the ability of Mobile Network Operators to manage and adjust training according to changing requirements.
Innovation Solution
Implementing consumer-controlled processes for Machine Learning model provisioning and training, allowing consumers to specify parameters such as learning architectures, data sources, and training status through interactions with NWDAF, enabling dynamic control over distributed and federated learning procedures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If distributed machine learning is implemented among multiple NWDAF instances, then training speed is accelerated and data privacy is protected, but consumer control over the training process is lost
Solution Approach 1:
The system segments control functions by introducing a consumer-facing interface that separates consumer control operations from the distributed training execution. The consumer can specify training parameters, select participating NWDAF instances, and control the training process through standardized service operations, while the actual distributed training is executed autonomously across multiple NWDAF instances. This segmentation resolves the contradiction by providing consumer control at the orchestration level while maintaining distributed execution for speed and privacy.
Solution Approach 2:
The patent introduces an intermediary control mechanism where the consumer interacts with the distributed ML system through standardized service operations (subscription, notification, control). This intermediary layer translates consumer control requirements into coordinated actions across multiple NWDAF instances without requiring direct consumer involvement in the complex distributed training processes, thus maintaining both consumer control and training efficiency.
2Reliability
If federated learning is used among multiple NWDAF instances, then data privacy is preserved, but the ability to dynamically adjust training according to changing requirements is limited
Solution Approach 1:
The system implements dynamic control capabilities where consumers can modify training parameters, add or remove participating NWDAF instances, and adjust training configurations during the training process. The service operations allow for dynamic subscription management, parameter updates, and process termination/restart, enabling the system to adapt to changing requirements while maintaining data privacy through the federated learning architecture.
Solution Approach 2:
The patent enables parameter changes during training by allowing consumers to update training configurations, specify different accuracy thresholds, modify time constraints, and adjust other hyperparameters without retraining from scratch. The system supports parameter modifications through service operations that maintain the federated learning structure while adapting to new requirements.
3Ease of operation
If centralized ML training is implemented, then consumer control is maintained, but data privacy and congestion issues arise
Solution Approach 1:
The system segments the training workload across multiple distributed NWDAF instances rather than centralizing it, while maintaining a centralized control interface for consumers. This segmentation allows control functions to remain centralized (maintaining ease of operation) while execution is distributed (protecting data privacy and reducing congestion).
Solution Approach 2:
The patent introduces an intermediary control layer that mediates between consumer control requirements and distributed execution. This intermediary maintains the appearance of centralized control for consumers while actually orchestrating distributed training across multiple instances, thus resolving the contradiction between consumer control and data privacy.
4Loss of time
If distributed resources are used for ML training, then training time is reduced, but complexity of managing multiple instances increases
Solution Approach 1:
The patent implements universal service operations that work across multiple NWDAF instances, providing a standardized interface for training management. The same service operations (subscription, notification, control) can be used regardless of the number or configuration of participating instances, simplifying management while maintaining distributed training benefits.
Solution Approach 2:
The intermediary control mechanism abstracts the complexity of managing multiple distributed instances from consumers. It handles instance selection, coordination, and synchronization internally, presenting a simplified interface to consumers who interact through standard service operations without needing to understand the underlying distributed system complexity.
Data Source
Figure 1~2
Figure 3~5
Figure 4
AI summary
A process is defined for consumer-controllable ML model provisioning and training in a wireless communication network. The process comprises two procedures, which respectively correspond to the two phases of an ML model provisioning process in, e.g., 5GC, i.e., the preparation/provisioning phase and the training execution phase. For the preparation/provisioning phase, new parameters are added to the request from the ML model consumer to the ML model generator (e.g., NWDAF), so that the latter can conduct the ML model provisioning according to the consumer requirements. For the training execution phase, interactions between the ML model consumer and generator(s) are considered, and the corresponding procedure for a consumer controlling the ML model training execution phase is defined.