Machine Learning Model Refinement via Feature Set Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional model refinement techniques for machine learning models are resource-intensive and non-scalable when refining for predetermined characteristics, requiring extensive time and effort due to feature perturbation and grouping strategies.
Innovation Solution
A method that captures model explanations, identifies feature sets, computes projected data sets, generates data set graphs to represent relationships, and validates these sets against target characteristic values to determine if they represent the predetermined characteristic, iteratively refining the model until compliance is achieved.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional model refinement techniques are used to refine machine learning models for predetermined characteristics, then the models can achieve desired predictive qualities, but the process becomes resource-intensive and non-scalable
Solution Approach 1:
The patent segments the model refinement process into distinct phases: generating multiple candidate models with different feature sets, evaluating them against predetermined characteristics, and iteratively selecting and refining promising candidates. This segmentation allows parallel processing of multiple models and reduces the computational burden on any single model, making the overall process more scalable and efficient while maintaining predictive quality.
Solution Approach 2:
The patent performs preliminary actions by generating multiple candidate models with different feature subsets before final model selection. This includes creating models with various feature combinations, evaluating their performance on predetermined characteristics in advance, and identifying promising candidates for further refinement. This preliminary exploration reduces the need for extensive iterative refinement of a single model, improving computational efficiency.
2Manufacturing precision
If feature perturbation and grouping strategies are applied to identify the best combination of features, then the models remain accurate and reflective of characteristics, but large investments of time are required
Solution Approach 1:
The patent applies partial action by generating multiple candidate models with different feature subsets rather than exhaustively searching all possible feature combinations. The system evaluates a selected number of candidate models with varying feature sets, identifies promising candidates based on preliminary performance metrics, and focuses refinement efforts on these selected models. This approach achieves sufficient feature selection accuracy without the prohibitive time cost of exhaustive search.
3Reliability
If extensive feature perturbation is performed to ensure model accuracy, then the models reflect predetermined characteristics, but the process becomes non-scalable
Solution Approach 1:
The patent segments the model development into multiple parallel candidate models, each with different feature subsets. This allows the system to evaluate multiple feature combinations simultaneously against predetermined characteristics, identifying models that accurately reflect desired characteristics without requiring extensive sequential perturbation of a single model. The segmented approach enables scaling to larger datasets and more complex characteristics.
Solution Approach 2:
The patent changes parameters by varying the feature subsets across multiple candidate models. Instead of extensively perturbing features in a single model, the system creates multiple models with different feature configurations, evaluates their performance on predetermined characteristics, and selects the best-performing models. This parameter variation approach maintains characteristic reflection accuracy while improving scalability.
Data Source
AI summary
A method for facilitating model refinement via evolutive computation for a predetermined characteristic is disclosed. The method includes capturing a model explanation for a model based on a sampled data set, the model explanation including a listing of model features based on feature scores; identifying feature sets based on the model explanation, the feature sets including a selection of the model features based on the listing; computing, by using raw data, projected data sets for each of the feature sets based on the corresponding selection; generating a data set graph based on the projected data sets, the data set graph representing a relationship between each of the projected data sets; selecting, by using the data set graph, the projected data sets; and validating the selected projected data sets based on a corresponding target characteristic value and a corresponding computed characteristic value.