Edge Server Deep Learning Model Retraining via Semi-Federated Data Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge server systems face inefficiencies in updating deep learning models due to lengthy data upload processes and inaccurate deployment caused by varying federated learning algorithms for different data scenarios, leading to suboptimal operational performance.
Innovation Solution
A semi-federated learning system with a data selector component and a federated learning algorithm selector component that generates new feature data and selects the best algorithm for retraining deep learning models, reducing data traffic and training time without uploading data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If new data is transmitted to a central server for re-training, then the software model can be updated, but the upload process takes a lengthy time
Solution Approach 1:
The patent segments the data processing workflow by introducing a data selector component that identifies and selects only the most relevant new data samples for re-training, rather than transmitting all collected data. This segmentation of data reduces the volume of data needing upload while maintaining update quality, directly addressing the time consumption issue.
Solution Approach 2:
The patent implements preliminary data selection and filtering actions at the edge server before data transmission to the central server. The data selector component performs preliminary processing to identify valuable data samples in advance, reducing the subsequent upload time and making the model update process more efficient.
2Productivity
If different federated learning algorithms are used for different data retrieval scenarios, then data retrieval can be optimized, but the deployment process becomes inaccurate
Solution Approach 1:
The patent employs parameter changes by dynamically selecting different federated learning algorithms based on data retrieval scenarios. The algorithm selector component adjusts algorithm parameters and selection based on characteristics of the retrieved data, enabling optimized data retrieval while maintaining deployment accuracy through adaptive algorithm selection.
3Reliability
If all collected data is uploaded for re-training, then the model can be comprehensively updated, but data traffic and training time increase
Solution Approach 1:
The patent extracts only the essential and most valuable data samples from the complete collected dataset for upload and re-training. The data selector component identifies and extracts high-quality data that is most relevant for model improvement, reducing data traffic consumption while maintaining update completeness by focusing on the most impactful data samples.
Data Source
AI summary
A system, method, and computer program product for implementing deep learning software model modification is provided. The method includes monitoring operational performance of a software model. An expected confidence level associated with the operational performance is first determined and it is determined that an inference associated with the expected confidence level is below a selected range of inferences associated with assigning new feature data as candidate video data. A candidate sequence comprising video data associated with the candidate video data is received and a similarity between frames of the candidate sequence is determined. A frame comprising a highest similarity with respect to segments of candidate video data is selected and it is detected that the frame is not associated with additional frames stored within a full cache structure. The software model is retrained such that the operational performance is modified.