Tree-Based Prediction Model Evaluation for Control Time Variability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for evaluating machine learning-based control prediction models are inefficient due to varying inference execution times, requiring repeated measurements to determine maximum execution time, which is time-consuming and labor-intensive.
Innovation Solution
A prediction system that includes a control operation unit, a predicted value acquisition unit, a prediction model generation unit based on a tree learning algorithm, and a prediction model evaluation unit, which calculates maximum execution time by comparing first and second execution times and processing counts, allowing for simpler evaluation of prediction models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning-based prediction models are used for control, then prediction accuracy is improved, but execution time becomes variable and difficult to determine
Solution Approach 1:
The patent applies preliminary action by pre-calculating execution times for multiple data types before actual control operations. The system measures and stores execution times for different data characteristics in advance, then retrieves these pre-measured values during runtime to determine maximum execution time without repeated measurements, thus resolving the time variability issue while maintaining prediction accuracy
Solution Approach 2:
The system performs preliminary classification of input data into multiple data types based on characteristics such as data range, distribution, or format. By categorizing data in advance and pre-measuring execution times for each category, the system can quickly determine expected execution time without repeated testing, thereby reducing time loss while preserving model accuracy
2Measurement precision
If repeated inference execution measurements are performed to determine maximum execution time, then execution time accuracy is improved, but evaluation time and effort increase
Solution Approach 1:
The patent applies copying by creating multiple copies of the prediction model and distributing them across different processing units or devices. Instead of repeatedly measuring execution time on a single model, the system runs multiple model copies in parallel with different input data, then aggregates the results to determine maximum execution time. This approach achieves accurate measurement without requiring sequential repeated testing, significantly reducing evaluation time and effort
Solution Approach 2:
The system performs preliminary classification of input data into multiple data types and pre-determines execution times for each data type category. By measuring execution time once per data type rather than repeatedly for each individual inference, the system achieves sufficient measurement accuracy while dramatically reducing the total evaluation time and manual effort required
Data Source
AI summary
A prediction system including a Central Processing Unit (CPU) and a memory storing a program. The CPU executes a control operation for controlling a control target, generates a prediction model based on a tree learning algorithm, acquires a predicted value by inputting, to the prediction model, a process value including one or more state values among state values that can be referenced by the CPU, acquires each of first and second execution times that are times taken for outputting the predicted value in response to input of first and second data to the prediction model, calculates a maximum execution time of the prediction model based on the first and second execution times, and evaluates the prediction model based on the maximum execution time.