Training Image Screening via Contrastive Model Parameter Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning model training techniques struggle with selecting appropriate training data, particularly when incorrect user-set categories and imbalanced data lead to ineffective training, and existing methods fail to identify inappropriate training images.
Innovation Solution
An image-for-training selecting apparatus and method that employs contrastive learning to train a first machine learning model with a first layer group, followed by a second model with a connected second layer group, calculating similarities between parameters before and after training to determine the appropriateness of training images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If training data is selected based on user-set correct categories, then training can proceed, but the training becomes ineffective when correct categories are incorrect or imbalanced
Solution Approach 1:
The patent implements a feedback mechanism where the system automatically evaluates training data quality by comparing model performance before and after training. The difference in accuracy serves as feedback to identify and exclude inappropriate training images, creating a closed-loop system that adapts to data quality issues without requiring manual verification of correct categories.
Solution Approach 2:
The system performs self-evaluation of training data appropriateness by automatically calculating accuracy differences and determining whether training images are suitable. This self-service capability eliminates the need for manual verification of correct categories, allowing the system to independently identify and filter out problematic training data.
2Manufacturing precision
If contrastive learning is used to train the first machine learning model, then feature extraction invariance is improved, but the complexity of the training process increases
Solution Approach 1:
The patent segments the training process into two distinct phases: first training with contrastive learning to establish invariant feature extraction, and second training to optimize classification accuracy. This segmentation allows each phase to focus on specific objectives, making the overall complex process more manageable and systematic.
Solution Approach 2:
The patent applies preliminary contrastive learning training before the main classification training. This preliminary action establishes a solid foundation of invariant feature representations, which then facilitates more effective subsequent training and improves overall model performance while managing complexity through staged development.
Data Source
AI summary
An image-for-training selecting apparatus for suitably selecting an image-for-training for training a machine learning model includes at least one processor executing: a first training process of training, by contrastive learning using an images-for-training set, a first machine learning model including a first layer group; a second training process of training a second machine learning model including the first layer group and a second layer group and employing the first machine learning model as a pre-trained model; a first calculating process of calculating a first similarity between a parameter of the first layer group after training by the first training process but before training by the second training process and a parameter of the first layer group after training by the second training process; and a first determining process of determining, based on the first similarity, whether the images-for-training set includes an inappropriate image-for-training.