Generation method, program, information processing apparatus, information processing method, and trained model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques face challenges in predicting with sufficient accuracy when the amount of labeled training data is insufficient, leading to suboptimal performance.
Innovation Solution
A method that combines labeled and unlabeled measurement data to generate a first trained model using supervised or semi-supervised learning, and subsequently generates a lighter second trained model for inference, leveraging data from similar machines to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only labeled training data is used for machine learning, then the model training is simple and fast, but the prediction accuracy is insufficient when labeled data is limited
Solution Approach 1:
The system performs self-service by automatically generating pseudo-labels for unlabeled data through the trained model, eliminating the need for manual labeling. The model infers labels for unlabeled measurement data, creating a self-sustaining training process that expands training data without external intervention.
Solution Approach 2:
The system performs preliminary action by pre-processing unlabeled data through initial training with labeled data, generating pseudo-labels before final model deployment. This preliminary labeling step prepares the data for subsequent fine-tuning and improves overall model accuracy.
2Measurement precision
If a large and complex trained model is generated to improve prediction accuracy, then the accuracy improves, but the model size increases and inference efficiency decreases
Solution Approach 1:
The training process is segmented into two distinct phases: first training with labeled data only, then fine-tuning with pseudo-labeled data. This segmentation allows the model to progressively improve accuracy while maintaining manageable complexity at each stage.
Solution Approach 2:
The system applies partial action by using only a portion of the available unlabeled data for pseudo-labeling, or by applying confidence thresholds to include only high-confidence predictions. This prevents overfitting while still benefiting from additional training data.
3Quantity of substance
If pseudo-labels are generated for all unlabeled data, then the training data quantity increases, but the reliability of training data decreases
Solution Approach 1:
The system changes parameters by adjusting confidence thresholds and selection criteria for pseudo-labeling. By dynamically setting thresholds based on model performance and data characteristics, the system optimizes the balance between quantity and reliability of training data.
Solution Approach 2:
The system implements feedback by evaluating the quality of pseudo-labeled data and adjusting the training process accordingly. High-confidence pseudo-labels are used for training, while low-confidence predictions are excluded or used for further verification, ensuring training data reliability.
Data Source
AI summary
This disclosure aims to provide a technique for improving the accuracy of prediction. An information processing apparatus performs a process including: acquiring a first data set related to a first machine, the first data set including: combined data, made by combining first measurement data with labels set for the first measurement data; and second measurement data for which no labels are set; generating a first trained model for inferring labels for measurement data of the first machine, based on the first data set; and generating a second trained model for inferring the labels for the measurement data of the first machine, based on the first trained model.