Collaborative Learning in Communication Networks Without Centralized Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collating data for machine-learning model training at a single network entity in 5G networks is slow, resource-intensive, and poses security risks, especially when dealing with sensitive data.
Innovation Solution
Implement a collaborative (federated) learning process where model training occurs locally at multiple network functions, with local updates combined at an aggregator entity to form a comprehensive model update, minimizing resource overhead and security risks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is collated at a single network entity for machine-learning model training, then comprehensive data collection is achieved, but latency increases and resource consumption increases
Solution Approach 1:
The patent segments the centralized data collection process into distributed local training operations at multiple network entities. Each entity trains models locally using its own data, eliminating the need to aggregate all data at a single point. This segmentation reduces latency while maintaining model training effectiveness through distributed computation.
2Measurement precision
If data is collated at a single network entity for machine-learning model training, then comprehensive data collection is achieved, but resource consumption increases
Solution Approach 1:
The patent divides the resource-intensive centralized training task into smaller local training operations distributed across multiple network entities. Each entity consumes resources locally for model training rather than concentrating all resource demands at a single entity, thereby reducing overall resource consumption while achieving comprehensive data utilization.
Solution Approach 2:
The patent uses model copies distributed to multiple network entities for local training. Instead of moving and processing all data at one location, model copies are replicated and trained locally at each entity, significantly reducing resource consumption at any single point while maintaining training comprehensiveness.
3Measurement precision
If data is collated at a single network entity for machine-learning model training, then comprehensive data collection is achieved, but security risks increase
Solution Approach 1:
The patent segments data storage and processing across multiple network entities rather than centralizing it. Each entity keeps its data locally and contributes only model updates or gradients to the collaborative training process, eliminating the security risks associated with data aggregation at a single entity while maintaining comprehensive data utilization for model training.
4Productivity
If collaborative learning is implemented across multiple network entities, then latency is reduced and resource consumption is minimized, but device complexity increases
Solution Approach 1:
The patent introduces a coordination entity that acts as an intermediary to manage the collaborative learning process across multiple network entities. This intermediary handles model distribution, collects local training updates, and coordinates aggregation, thereby simplifying the complexity for individual entities while enabling efficient distributed training across the network.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In one aspect, a method performed by a network entity in a communications network is provided. The method comprises obtaining identification information for a plurality of candidate network entities in the communications network, in which the identification information indicates that each of the candidate network entities is configured to participate in collaborative learning. The method further comprises sending a request for each of the candidate network entities, in which the request comprises one or more selection criteria, and receiving one or more response messages comprising an indication of which of the candidate network entities satisfy the one or more selection criteria. Based on the indication in the one or more response messages, the network entity selects one or more of the plurality of candidate network entities to participate in a collaborative learning process to train a model using a machine learning algorithm.