Synthetic Data Generation With Multi-Model Quality Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for generating synthetic data are limited in their ability to closely mimic real data, which is essential for effective machine learning model training and decision-making.
Innovation Solution
A method involving multiple trained machine learning models is used to generate synthetic data vectors by computing error functions based on similarity, prediction, and bias assessment errors, iteratively updating hyperparameters to optimize the synthetic data generation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current techniques are used to generate synthetic data, then the generation process is simple, but the synthetic data does not closely mimic real data
Solution Approach 1:
The patent divides the synthetic data generation process into multiple stages using different machine learning models: a first model generates initial synthetic data vectors, a second model refines them by comparing with real data, and a third model evaluates quality metrics. This segmentation allows each model to focus on specific aspects of data similarity, achieving high measurement precision while maintaining manageable process complexity through modular architecture.
Solution Approach 2:
The patent implements feedback mechanisms where the second machine learning model uses real data to refine synthetic data vectors, and the third model provides quality evaluation metrics. This feedback loop enables continuous improvement of data similarity to real data distributions, directly addressing the contradiction between simplicity and precision by using iterative refinement.
2Measurement precision
If multiple machine learning models are used to generate synthetic data, then the accuracy of synthetic data is improved, but the computational resources required increase
Solution Approach 1:
The patent applies partial action by using different models for different purposes: the first model performs initial generation with moderate computational resources, the second model applies targeted refinement only where needed by comparing with real data, and the third model provides selective quality evaluation. This partial application of computational resources across multiple models achieves high accuracy while optimizing energy consumption compared to using a single complex model.
3Adaptability or versatility
If synthetic data is generated to mimic real data properties, then the usefulness for training models is improved, but the difficulty of ensuring data quality increases
Solution Approach 1:
The patent introduces a third machine learning model as an intermediary evaluator that assesses the quality of synthetic data vectors against real data. This intermediary quality assessment mechanism provides measurable metrics for data quality, making it easier to detect and measure whether the synthetic data adequately mimics real data properties while maintaining versatility for training purposes.
Data Source
AI summary
A system and method include generating synthetic data by generating a first set of hyperparameters for a first trained machine learning model and a second set of hyperparameters for a second trained machine learning model, generating a plurality of synthetic data vectors using the first and second trained machine learning models, computing an error function for the first and second set of hyperparameters using a third machine learning model, computing an objective function value, responsive to determining that the objective function value is not an optimal value, updating the first set of hyperparameters and the second set of hyperparameters or responsive to determining that the objective function value is an optimal value outputting the plurality of synthetic data vectors as a set of synthetic data.