Machine Learning Operations Control with Local Device Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The massive transfer of training data for machine learning algorithms in cellular networks can severely impact network performance and spectral efficiency, particularly when training is conducted on the network side, leading to potential congestion through interfaces.
Innovation Solution
User devices are equipped to perform machine learning operations control by receiving measurement configurations, obtaining and comparing measurement values to acceptance conditions, and training machine learning models for radio resource management, with trained models transmitted to access nodes or used for internal operations, thereby reducing the need for extensive data transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning training is conducted on the network side, then centralized control and model management is improved, but network congestion and spectral efficiency deteriorate due to massive data transfer
Solution Approach 1:
The patent segments the machine learning training process by distributing it across multiple user devices rather than centralizing it in the network. Each user device independently performs training operations using local measurement data, eliminating the need for massive data transfer to the network while maintaining decentralized model development.
Solution Approach 2:
Instead of the conventional approach where data is collected and processed centrally in the network, the patent inverts the processing location to the user devices. The measurement configuration is provided by the network, but the actual training computation is performed locally at user devices, reversing the traditional data flow direction.
2Measurement precision
If machine learning training data is transferred to the network, then centralized model training is improved, but radio resource consumption and network congestion worsen
Solution Approach 1:
The patent extracts the training computation function from the network and relocates it to user devices. By taking out the data transfer requirement and performing training locally using only the measurement configuration provided by the network, the solution eliminates massive data transfer while preserving model training capability.
Solution Approach 2:
User devices perform self-service by conducting their own training operations using local resources and the measurement configuration provided by the network. Each device independently processes its measurement data without requiring network-side processing, thereby eliminating the need for data transfer while maintaining training functionality.
3Loss of energy
If user devices perform local training, then radio resource efficiency is improved, but device complexity and processing requirements worsen
Solution Approach 1:
The patent applies partial action by having user devices perform only the specific training operations needed for their local context using the measurement configuration provided by the network. Rather than requiring full-featured training capabilities, devices perform targeted training computations that are sufficient for their specific radio resource management needs.
Data Source
AI summary
As an aspect, there is provided an apparatus, caused at least to: receive, by a user device from an access node, a measurement configuration for training of a machine learning operations control model, the measurement configuration comprising an indication of at least one parameter to be measured and at least one acceptance condition for measurement values; obtain at least one measurement value; compare the at least one measurement value to the at least one acceptance condition for defining usability of the at least one measurement value for the training, and when the at least one measurement value is usable for the training, carry out the training of the operations control model, and when the at least one measurement value is not usable for the training, transmit the at least one measurement value to the access node.