Synthetic Data Generation via Trained ML Model Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional methods of synthetic data generation often result in biased or inaccurate results due to sampling from existing datasets not providing a representative sample of the population.

Innovation Solution

A method using trained machine learning models to generate synthetic data, where each model is trained to generate synthetic data and associated with a target value of an evaluation function, allowing for the selection of the most suitable model based on user requests.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If sampling from existing datasets is used for synthetic data generation, then the process is simple and fast, but the results are biased and inaccurate due to non-representative sampling

Engineering Contradiction:
Improvedata generation speedVSAvoiddata accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent uses trained machine learning models to copy and reproduce the statistical properties and patterns of real-world data distributions, rather than简单地 sampling from existing datasets. This allows accurate replication of data characteristics while enabling flexible generation of representative synthetic data.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system allows users to specify target values for evaluation functions (such as statistical properties, distribution characteristics, or quality metrics), and the model selection mechanism chooses models trained to achieve these specific parameter targets, thereby controlling the accuracy and representativeness of generated data.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If multiple trained ML models are maintained with different target values, then the system can fulfill diverse user requests accurately, but the device complexity increases

Engineering Contradiction:
Improverequest fulfillment capabilityVSAvoidmodel management complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system maintains a collection of ML models, each optimized for specific target values or data characteristics. The model selection mechanism acts as a universal interface that routes user requests to the appropriate specialized model, allowing the system to handle diverse requirements while managing complexity through organized model curation.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The model selection mechanism serves as an intermediary layer between user requests and the collection of trained models. It evaluates user requirements against model capabilities and target values, selecting the most appropriate model without requiring users to directly manage or understand the complexity of multiple models.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Ease of manufacture

If traditional sampling methods are used, then the implementation is straightforward, but the synthetic data does not provide a representative sample of the population

Engineering Contradiction:
Improveimplementation simplicityVSAvoiddata representativeness
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

Machine learning models are pre-trained on real-world datasets to learn accurate data distributions, statistical properties, and patterns before deployment. This preliminary training action ensures that when the models generate synthetic data, they inherently produce representative samples without requiring complex sampling strategies at generation time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional mechanical sampling methods (random sampling, stratified sampling, etc.) with intelligent ML-based generation. Instead of using algorithmic sampling rules that may fail to capture complex data relationships, the system uses trained models that have learned the underlying data structure, providing more reliable representativeness.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4546225A1Generating synthetic data with trained ML models
Publication Date: 2025.04.30 BUNDESDRUCKEREI GMBH
  • EP4546225A1 patent drawingFigure 1
  • EP4546225A1 patent drawingFigure 2
  • EP4546225A1 patent drawingFigure 3

AI summary

Disclosed is a method for generation of synthetic data. The method comprises the following steps: providing a set of one or more trained ML models, each trained ML model being trained to generate synthetic data, wherein each trained ML model of the set of one or more trained ML models is associated with a respective target value of an evaluation function that has been used to evaluate the set of one or more ML models for the training; receiving from a user a request to generate synthetic data; using the request and the target value(s) for selecting a trained ML model of the set of trained ML models that fulfills the request; using the selected ML model for generating synthetic data.