Machine Learning Model Generation with Fixed Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models face accuracy deterioration due to changes in input data distribution over time, and current techniques like the T2 statistic are impractical for detecting such changes, especially in high-dimensional data, leading to loss of important information and increased computational costs.
Innovation Solution
A model generation program and device that acquires results from a first machine learning model, selects relevant data, and generates new machine learning models with fixed parameters to detect accuracy deterioration by reusing trained parameters and narrowing model application areas, reducing computational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained with all available data, then model accuracy is improved, but computational costs increase
Solution Approach 1:
The patent extracts and removes unnecessary data from the training set by identifying and eliminating redundant samples. The system analyzes data characteristics and selectively removes data points that do not contribute to model accuracy, thereby reducing computational costs while maintaining predictive performance.
Solution Approach 2:
The patent changes parameters by adjusting the selection criteria for training data based on data characteristics. The system dynamically modifies which data points are included in training by analyzing features such as data distribution, redundancy, and importance, thereby optimizing the balance between accuracy and computational efficiency.
2Measurement precision
If machine learning models are retrained frequently to adapt to data distribution changes, then model accuracy is maintained, but computational costs and time increase
Solution Approach 1:
The patent performs preliminary analysis of data distribution changes before full retraining is necessary. The system monitors data characteristics and identifies when changes warrant model updates, allowing for proactive adjustments that maintain accuracy while avoiding unnecessary frequent retraining cycles.
Solution Approach 2:
The patent applies partial retraining by updating only specific components of the model or using a subset of relevant data rather than complete retraining. This selective approach maintains model accuracy in response to data distribution changes while significantly reducing the time and computational resources required compared to full retraining.
3Loss of energy
If important features are reduced in high-dimensional data, then computational costs are decreased, but information loss occurs
Solution Approach 1:
The patent applies local quality by treating different features differently based on their importance and contribution to the model. Rather than uniformly reducing all features, the system identifies and preserves locally important features while reducing or eliminating less significant ones, thereby maintaining information quality while reducing computational burden.
Solution Approach 2:
The patent changes parameters by dynamically adjusting feature selection based on data characteristics and model performance. The system modifies which features are included in training by analyzing their importance, correlation, and contribution to predictive accuracy, optimizing the balance between information retention and computational efficiency.
4Measurement precision
If the model application area is narrowed to improve precision, then measurement precision is improved, but adaptability decreases
Solution Approach 1:
The patent makes the model application area dynamic rather than fixed. The system adjusts the scope and boundaries of the model's application based on changing data characteristics and performance requirements, allowing the model to adapt its precision level to different operational contexts while maintaining overall effectiveness.
Solution Approach 2:
The patent changes parameters by adjusting the model's application scope and decision thresholds based on data characteristics. The system dynamically modifies parameters such as confidence thresholds, feature weights, and application boundaries to optimize precision for specific tasks while retaining the ability to adapt to different scenarios and data distributions.
Data Source
AI summary
A non-transitory computer-readable recording medium stores a model generation program for causing a computer to execute a process including: acquiring results output from a first machine learning model in response to input of a first plurality of pieces of data to the first machine learning model; selecting a second plurality of pieces of data from the first plurality of pieces of data, based on the results; and generating a second machine learning model by executing machine learning of the first machine learning model, by using the second plurality of pieces of data as input, with some parameters being fixed among a plurality of parameters included in the first machine learning model.