Online Neural Network Retraining via Intermediary Training Environment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deployed neural networks face challenges in improving their performance without disrupting critical functions, as testing or upgrading them is impractical due to resource constraints and operational limitations.
Innovation Solution
A method for online training of deployed neural networks involves using a more complex neural network model on a secondary computing system to assess and retrain the initial model, leveraging collected data to improve accuracy and deploy updates without significant disruptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deployed neural network model is retrained or upgraded to improve accuracy, then the model performance is improved, but the system disruption and operational impact increase
Solution Approach 1:
The system divides the neural network training process into separate environments: a production environment for critical operations and a training environment for model improvement. This segmentation allows training activities to occur without disrupting production operations, resolving the contradiction between improving accuracy and maintaining operational stability.
Solution Approach 2:
The patent introduces an intermediary training environment that acts as a mediator between the production system and the training process. This intermediary environment collects data from production, performs training there, and then deploys updated models back to production, thereby improving accuracy without direct disruption to operational systems.
2Measurement precision
If computing resources are allocated for training and testing the neural network, then the model accuracy can be improved, but the resource consumption and system complexity increase
Solution Approach 1:
The training environment is designed to serve multiple functions: collecting data from production systems, performing training operations, evaluating model performance, and deploying updated models. This multi-functionality reduces the need for separate dedicated systems, thereby improving model accuracy without proportionally increasing computing system complexity.
Solution Approach 2:
The system enables self-service training where the production environment automatically provides data to the training environment, and the training environment automatically evaluates and deploys models back to production. This automated self-service process reduces manual intervention and system complexity while maintaining the ability to improve model accuracy.
3Reliability
If the neural network operates in a critical function, then the system provides essential services, but the ability to test or upgrade the model is limited
Solution Approach 1:
The system performs all training, testing, and validation actions preliminarily in the training environment before deploying changes to the production system. This preliminary action allows extensive testing and upgrading to occur in advance, ensuring that critical functions maintain their reliability while still enabling model improvement through the staged deployment process.
Data Source
AI summary
Embodiments presented herein facilitate improvement of a deployed neural network model's accuracy without significantly affecting its operation. In one or more embodiments, online training of the deployed model may be performed using a second neural network model that has higher accuracy than the deployed neural network model. In one or more embodiments, the second neural network model may also be improved online. Embodiments may be deployed in system, such as edge computing environments, in which neural networks deployed at the edge can be centrally monitored and updated.