Continual Learning Defect Diagnosis With Buffer Data Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based radiographic testing methods face issues with catastrophic forgetting, leading to reduced accuracy due to the need to relearn all previous data sets, which results in excessive storage requirements.
Innovation Solution
A method and device for defect diagnosis using continual learning, involving training with a buffer training data set and current data set, calculating influence and change, and updating the buffer data set based on plasticity and stability scores to maintain accuracy while minimizing storage needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If transfer learning is used to relearn features of new data sets, then the model adapts to new data, but catastrophic forgetting occurs and accuracy of previous data is reduced
Solution Approach 1:
The patent stores important historical data in a buffer before it can be forgotten. By proactively saving representative samples from previous data sets in the buffer, the system ensures that critical information is preserved and can be relearned when needed, preventing catastrophic forgetting before it occurs.
Solution Approach 2:
The buffer acts as an intermediary between the model and historical data. Instead of directly accessing all historical data or risking complete forgetting, the buffer serves as a mediator that stores and manages a curated subset of important data, allowing the model to selectively relearn from this intermediate storage without being overwhelmed by the full historical data set.
2Reliability
If all previous training data is relearned to prevent catastrophic forgetting, then accuracy of previous data is maintained, but storage space requirements increase excessively
Solution Approach 1:
The patent extracts only the most important and representative data points from historical data sets and stores them in the buffer. Instead of retaining all previous training data, the system selectively extracts critical samples that capture the essential features and patterns, significantly reducing storage requirements while maintaining the ability to prevent catastrophic forgetting.
Solution Approach 2:
The system changes the parameter of data retention from storing complete data sets to storing a curated subset based on importance criteria. By transforming the storage approach from comprehensive to selective, the system achieves the same protective effect against catastrophic forgetting with much lower storage costs.
3Volume of stationary object
If a buffer is used to store selected training data, then storage space is reduced, but selecting and managing the buffer data set adds system complexity
Solution Approach 1:
The buffer management system operates autonomously by automatically evaluating the importance of data points and making decisions about what to store and when to clear the buffer. The system serves itself by implementing self-contained mechanisms for data selection, importance assessment, and buffer maintenance without requiring complex external management infrastructure.
Solution Approach 2:
The patent transforms the complex problem of buffer management into a simpler parameter-based decision system. By defining clear importance criteria and threshold parameters, the system converts complex judgment tasks into straightforward parameter comparisons, reducing the overall system complexity while maintaining effective buffer management.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Proposed are a device for diagnosing a defect using a model based on continual learning and a method therefor, and the method for diagnosing the defect includes a step of loading, by a training unit, a buffer training data set, which is training data selected according to a degree of influence on prediction performance of a current model from among a past training data set in the continual learning, a step of training the current model using the buffer training data set and a current training data set, and a step of detecting, by a detection unit, the defect in a radiographic image using the current model when the radiographic image is input.