Statistical-Deviation Classification for Sensor Changes Without Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models require retraining when changes occur in the target object or measurement system, such as sensor replacement, leading to inefficiencies and additional data processing.
Innovation Solution
A method using a trained model that calculates statistical deviations in data to classify new data into different classifications by recording numerical interval data, allowing reuse of the model without retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning model is retrained when changes occur in the target object or sensor, then the model accuracy is maintained, but the time loss and processing complexity increase
Solution Approach 1:
The patent calculates reference calculation data in advance for multiple classification types using the trained model, storing these as numerical interval data before actual classification is needed. This preliminary preparation allows the system to handle sensor replacements or target object changes without retraining, as the reference data is already computed and stored for quick comparison during runtime classification operations.
2Adaptability or versatility
If a machine learning model is retrained when changes occur in the target object or sensor, then the model adapts to new conditions, but the device complexity and processing requirements increase
Solution Approach 1:
The patent extracts the essential classification capability from the complex retraining process by calculating reference calculation data separately for each classification type and storing it as numerical interval data. This extraction allows the system to adapt to changes by simply comparing new data against the pre-stored reference intervals rather than performing complex model retraining, significantly reducing processing complexity while maintaining adaptability.
Solution Approach 2:
Instead of creating a new trained model when changes occur, the patent creates a copy of the classification process by pre-calculating reference calculation data for all possible classification types and storing them as numerical interval data. This copying approach allows the system to handle new target objects or sensors by comparing against the copied reference data, avoiding the complexity of model retraining while maintaining full adaptability.
3Reliability
If extensive data processing is performed for retraining, then the model performance is optimized, but the loss of time and computational resources increase
Solution Approach 1:
The patent performs the computationally intensive calculation of reference calculation data in advance during system setup, storing the results as numerical interval data. This preliminary action consolidates the computational workload into a one-time process rather than repeated retraining operations, significantly reducing energy consumption and computational resource loss during subsequent classification tasks, even when target objects or sensors change.
Data Source
AI summary
A classification method includes calculating a plurality of pieces of reference calculation data from a plurality of pieces of data belonging to a second classification by using a machine learning model that has learned learning data including a plurality of pieces of data belonging to a first classification, calculating classification calculation data from one piece of acquisition data by using the machine learning model, and comparing data based on the plurality of pieces of reference calculation data with the classification calculation data to classify whether the one piece of acquisition data belongs to the second classification. The plurality of pieces of data belonging to the first classification, the plurality of pieces of data belonging to the second classification, and the one piece of acquisition data have a statistical deviation, and the second classification is different from a classification of the plurality of pieces of data included in the learning data.