Synthetic Data Generation for Privacy-Compliant Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data protection and privacy regulations restrict the use of real data for training artificial intelligence and machine learning models, making it challenging to create effective training data while ensuring compliance.
Innovation Solution
A computer-implemented process that accesses static and dynamic system data to derive relevant data, using distance-based and importance-based metrics, generates synthetic data that complies with privacy regulations, and is used to train machine learning models for classification purposes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real data is used for training machine learning models, then model training effectiveness is improved, but data protection and privacy compliance deteriorates
Solution Approach 1:
The patent creates synthetic data that copies the statistical properties, distributions, and relationships of real data without containing actual sensitive information. This allows machine learning models to be trained on data that mimics real-world patterns while ensuring privacy compliance, as the synthetic data contains no personally identifiable information
Solution Approach 2:
The patent introduces synthetic data as an intermediary between real data and machine learning models. This intermediary preserves the essential characteristics needed for model training while eliminating privacy risks, allowing the model to learn from data that is statistically representative but legally safe to use
2Object-affected harmful factors
If synthetic data is generated to comply with privacy regulations, then data protection compliance is improved, but model training effectiveness may deteriorate
Solution Approach 1:
The patent carefully adjusts parameters of the synthetic data generation process to preserve critical statistical properties such as data distributions, correlations, and relationships. By maintaining these parameters while removing sensitive information, the synthetic data remains effective for model training while ensuring privacy compliance
3Measurement precision
If manual data selection and exploration is performed, then data relevance accuracy is improved, but productivity deteriorates
Solution Approach 1:
The patent implements automated systems that self-select and explore data without requiring manual intervention. The system automatically identifies relevant data, explores its properties, and generates synthetic data, eliminating the time-consuming manual processes while maintaining accuracy through algorithmic precision
4Manufacturing precision
If comprehensive data exploration is performed to understand data structures and correlations, then synthetic data quality is improved, but time consumption increases
Solution Approach 1:
The patent performs comprehensive data exploration and analysis as a preliminary step before synthetic data generation. By understanding data structures, distributions, and correlations upfront, the system can efficiently generate high-quality synthetic data without needing to revisit exploration phases, reducing overall time consumption while maintaining quality
Data Source
AI summary
Static and dynamic process data of a system are accessed. Thereafter, using this accessed process data, a subset of such data forming relevant data for a particular context is derived. The data is then explored using a computer-implemented process or processes to automatically get insight into information about structures, distributions and correlations of the relevant data. Rules can be generated based on the exploring of relevant data that describe data dependencies within the relevant data. These generated rules can later be used to generate synthetic data. Such synthetic data, in turn, can be used to for a variety of purposes including the training of a machine learning model while, at the same time, complying with applicable privacy and data protection laws and regulations.