Active Learning Data Filtering for 5G Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In 5G communication systems, the efficient re-training of machine learning models is hindered by the high computational effort required for data labelling and processing, especially in environments with limited resources, where transferring all inference data for active learning is not feasible due to processing power constraints and data volume.
Innovation Solution
Implementing active learning techniques to selectively filter and extract the most useful data samples from inference data for re-training, using a distributed query function between inference and training entities to reduce the number of samples needed and optimize data transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all inference data is transferred for active learning re-training, then model performance can be improved, but computational effort and data transfer load become prohibitively high
Solution Approach 1:
The patent segments the inference data by identifying and separating the most useful data samples from the complete inference dataset. Instead of transferring all data, only a segmented subset containing the most valuable training samples is selected and transferred for re-training, thereby reducing computational effort while maintaining model performance improvement.
Solution Approach 2:
The patent extracts the most useful data samples from the complete inference dataset by applying selection criteria and filtering mechanisms. This extraction process identifies and removes only the essential samples needed for effective re-training, eliminating the need to process or transfer the entire dataset and thus reducing the computational burden.
2Measurement precision
If all inference data is transferred for re-training, then training accuracy can be improved, but data transfer load increases beyond feasible limits
Solution Approach 1:
The patent extracts only the most useful data samples from the complete inference dataset based on predefined selection criteria. This extraction mechanism identifies and isolates the essential samples required for accurate re-training, thereby maintaining training accuracy while dramatically reducing the volume of data that needs to be transferred.
Solution Approach 2:
The patent applies partial action by transferring only a portion of the inference data - specifically, the most useful samples - rather than the complete dataset. This partial transfer approach provides sufficient training material to achieve accurate model re-training without the excessive data transfer load that would result from transmitting all inference data.
3Productivity
If comprehensive data processing is performed for active learning, then model convergence speed can be improved, but processing time and resources increase
Solution Approach 1:
The patent segments the data processing task by focusing computational resources only on identifying and processing the most useful data samples rather than the entire inference dataset. This segmentation allows the active learning process to converge faster by working with a smaller, more relevant subset of data, thereby reducing processing time while maintaining convergence speed.
Solution Approach 2:
The patent applies partial action by performing comprehensive data processing only on the selected useful samples rather than on all inference data. This approach achieves effective model convergence with reduced processing time and computational resources, as the processing is concentrated on the subset of data that provides the most value for learning.
Data Source
AI summary
There is disclosed a method. The method comprises, in response to a request from an inference consumer for inference data samples, obtaining inference data samples using a machine learning model. The method further comprises querying the obtained data samples to generate a filtered set of data samples. The method further comprises sending the filtered set of data samples to a training entity for training the machine learning model. The method further comprises sending the obtained inference data samples to the inference consumer.