Training Data Generation Using ML Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating training data, especially for abnormal states, often result in inaccurate teacher labels when comparing input data, leading to deteriorated learning model accuracy due to missed important features during prediction tasks.
Innovation Solution
A method involving an information processing apparatus that acquires features from a machine learning model for both chronological states before and after a predetermined action, generates new training data by changing the teacher label based on these features, and updates the learning model to improve accuracy by using these new data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training data are generated by giving actual result as teacher label to input data, then training data can be collected for abnormal states, but accuracy of teacher label assignment deteriorates due to missed important features
Solution Approach 1:
The patent applies preliminary action by acquiring features from the machine learning model before finalizing the teacher label assignment. Specifically, the system acquires a first feature from the model's prediction about the target state, then uses this feature information to generate training data with corrected teacher labels. This preliminary feature extraction allows the system to identify important features before label assignment, preventing feature omission in the training data generation process.
2Ease of manufacture
If existing methods are used for generating training data, then the process is simple, but learning model accuracy deteriorates due to inaccurate teacher labels
Solution Approach 1:
The patent implements feedback by using the machine learning model's own predictions to generate training data. The system acquires features from the model's prediction results and uses these features to create training data with corrected teacher labels. This feedback mechanism ensures that the training data generation process leverages the model's learned patterns, improving teacher label accuracy and subsequently enhancing learning model accuracy without significantly complicating the overall process.
3Productivity
If teacher label is assigned without considering model features, then the process is fast, but important features for prediction task are missed
Solution Approach 1:
The patent applies preliminary action by extracting features from the machine learning model before finalizing teacher label assignment. The system acquires features representing the target state prediction and uses these pre-extracted features to generate training data. This ensures that important feature information is captured and included in the training data, preventing feature loss while maintaining efficient processing speeds.
Data Source
AI summary
A non-transitory computer-readable recording medium has stored therein a program that causes a computer to execute a process comprising: acquiring a first feature from a machine learning model that estimates a first result of a target after a first period in response to an input of a first chronological state of the target in the first period, the first feature being a feature of the first chronological state; acquiring a second feature by inputting a second chronological state to the machine learning model, the second feature being a feature of the second chronological state in a second period including a period after the first result is determined; and generating, based on the first feature and the second feature, training data that takes the second chronological state as an explanatory variable and takes a second result as an objective variable, the second result being obtained by changing the determined first result.