Machine Learning Model Training Using Selected Time-Series Segments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data-driven decision-making processes are often time-consuming and require large datasets for training machine learning models, which can be impractical in applications with limited data availability and the need for rapid response times.
Innovation Solution
A procedure for training machine learning models that selects specific parts of data combined with classification data units, allowing for improved decision-making quality, reduced data requirements, and automated processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning training uses large datasets with thousands of data sets, then model accuracy is improved, but training time becomes excessively long and the process becomes impractical for rapid decision-making
Solution Approach 1:
The patent extracts and utilizes domain knowledge and expert rules to pre-process and filter training data, removing the need to train on all raw data. This extraction of essential patterns allows the model to achieve high accuracy with significantly fewer training examples, directly resolving the contradiction between accuracy and training time.
Solution Approach 2:
The system performs preliminary actions by pre-defining classification categories, expert rules, and data filtering criteria before training begins. This preliminary structuring of knowledge allows the machine learning model to focus learning on fewer, more relevant examples, reducing training time while maintaining accuracy.
2Reliability
If comprehensive datasets are used for training, then model reliability is improved, but data availability becomes limited in many practical applications
Solution Approach 1:
The patent changes the parameters of the training approach by incorporating domain-specific knowledge and expert rules as additional training signals. This allows the model to achieve reliable performance with smaller datasets by augmenting limited data with structured domain knowledge, effectively resolving the contradiction between reliability and data quantity.
Solution Approach 2:
The system creates a composite training approach that combines limited empirical data with structured domain knowledge and expert rules. This composite training methodology enables the model to achieve reliability comparable to training on large datasets, even when data availability is limited.
3Measurement precision
If specialized analysis tools are developed for monitoring systems, then measurement precision is improved, but device complexity and adaptation effort increase significantly
Solution Approach 1:
The patent implements a universal machine learning framework that can handle multiple monitoring tasks and decision-making scenarios through a single system. By using general-purpose machine learning algorithms combined with domain knowledge, the system achieves specialized monitoring precision without requiring separate complex tools for each application, thus resolving the contradiction between precision and complexity.
4Adaptability or versatility
If machine learning models are trained extensively, then adaptability to different use cases is improved, but training resources and time requirements increase
Solution Approach 1:
The patent segments the training process into distinct phases: data preprocessing with domain knowledge, model training on filtered data, and deployment. This segmentation allows the model to be efficiently trained for specific applications without requiring extensive retraining, achieving adaptability while reducing training time through the modular approach.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method for training machine learning models (51), having the steps of: detecting (S10) data (70) in the form of time series data using one or more computers (52), said data being obtained by means of one or more measuring devices (60-62), in each case in the form of a sensor for measuring a physical variable; receiving (S12) multiple classification data units relating to the data (70) using the one or more computers (52); receiving (S13) a selected part (71) of the data (70) using the one or more computers (52) for each of the classification data units; and training (S14) multiple machine learning models (51) using the one or more computers (52), in each case on the basis of at least one of the classification data units and the at least one corresponding selected part (71) of the data, wherein the multiple machine learning models (51) represent multiple instances of the same machine learning model.