AI Model Updating via Automatic Labeling and Iterative Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for updating artificial intelligence models are inefficient, requiring significant time and manpower, and are not well-suited to handle changes in industrial infrastructure facilities or environments over time, leading to decreased prediction accuracy.
Innovation Solution
A method and device for automatically updating artificial intelligence models by collecting new datasets, selecting update datasets, generating labels based on model outputs, and repeatedly refining the update dataset until performance convergence, thereby maintaining high prediction accuracy despite changes in industrial infrastructure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual labeling and model building tasks are performed by people, then the model can be initially built, but a lot of time and manpower have to be invested
Solution Approach 1:
The system performs automatic labeling by having the AI model itself generate labels for new data without human intervention. The model uses its own trained capabilities to annotate data, effectively serving itself in the labeling task that would otherwise require human resources.
Solution Approach 2:
The patent replaces manual mechanical labeling processes with an automated computational system. Instead of humans manually annotating data, the system automatically processes data through computational algorithms and model outputs to generate labels, substituting human labor with automated mechanical processing.
2Reliability
If the AI model is designed according to initial facility environment, then the model can be built initially, but prediction accuracy gradually decreases over time due to facility changes
Solution Approach 1:
The system continuously collects new data from the facility environment and continuously updates the AI model through automated labeling and retraining processes. This continuous action ensures the model adapts to changing facility conditions over time, maintaining high prediction accuracy without interruption.
Solution Approach 2:
The system uses new data collected from the facility environment as feedback to update and improve the model. By continuously incorporating new environmental data through automated labeling and model retraining, the system adjusts to facility changes and maintains optimal performance.
3Productivity
If automatic labeling is implemented, then time and manpower for labeling can be reduced, but the process of selecting update dataset and generating labels must be repeated until convergence
Solution Approach 1:
The iterative process uses performance metrics as feedback to determine when convergence is achieved. The system automatically evaluates model performance after each training cycle and uses this feedback to decide whether to continue iterating or stop, managing the complexity through automated decision-making based on performance thresholds.
Data Source
AI summary
An artificial intelligence model updating method includes collecting a new dataset generated by performing at least one task using an artificial intelligence model, selecting an update dataset from entire dataset consisting of the initial dataset and multiple unit data of the new dataset, and generating a label of the update dataset based on an output of the artificial intelligence model trained using the initial dataset, wherein a process of reselecting the update dataset until the artificial intelligence model performance is converged and generating a label of the reselected update dataset is performed repeatedly, and thus, automatic labeling with an accuracy almost comparable to manual labeling performed by human may be provided.