Home Arrow Icon Knowledge base Arrow Icon Global Arrow Icon How does APE maintain referential integrity in synthetic data


How does APE maintain referential integrity in synthetic data


APE (Automatic Post-Editing) maintains referential integrity in synthetic data generation through advanced techniques that ensure the relationships among different data elements are preserved. This is crucial, especially when dealing with relational databases where data integrity is enforced through foreign keys and other constraints.

Methods for Maintaining Referential Integrity

1. Hierarchical Generative Adversarial Networks (GANs): APE utilizes hierarchical GANs to synthesize data while preserving referential integrity. This approach involves generating data at multiple granularity levels, ensuring that foreign key relationships between tables are maintained. For example, when synthesizing customer and order data, the GAN can generate customer information first and then conditionally generate order details based on the generated customer data, thus preserving the necessary relationships[1].

2. Data Clustering and Synthesis: By employing unsupervised machine learning techniques, APE can cluster data at a parent level (e.g., customers) before synthesizing related child data (e.g., orders). This method ensures that the synthetic data reflects the hierarchical structure of the original database, allowing for accurate foreign key relationships to be established[1].

3. Dynamic Data Modeling: APE can adapt to changes in the database schema by using intelligent automation to redefine relationships between data tables. This flexibility allows for ongoing maintenance of referential integrity even as the underlying data structures evolve[4].

4. Validation Techniques: Post-generation validation is employed to ensure that the synthetic data adheres to the original data's structural integrity. This includes checking that all foreign keys in the synthetic dataset correctly reference existing primary keys, thereby ensuring that the relationships among data elements are accurate and intact[2].

5. Statistical Preservation: APE focuses on maintaining the statistical properties of the original dataset, including distributions and correlations among data points. By ensuring that these properties are preserved, the generated synthetic data remains representative of the real data, which inherently supports referential integrity[5].

Conclusion

Through the use of hierarchical GANs, data clustering, dynamic modeling, and rigorous validation techniques, APE effectively maintains referential integrity in synthetic data generation. This capability not only enhances the utility of the synthetic data for various applications but also ensures compliance with data integrity standards essential for relational databases.

Citations:
[1] https://hazy.com/resources/2020/04/27/generating-synthetic-data-with-referential-integrity-using-gans
[2] https://www.fca.org.uk/publications/research-articles/exploring-synthetic-data-validation-privacy-utility-fidelity
[3] https://www.genrocket.com/synthetic-data-generation/
[4] https://www.genrocket.com/blog/data-modeling-and-referential-integrity/
[5] https://gretel.ai/what-is/synthetic-data-generation
[6] https://aclanthology.org/2022.lrec-1.93/
[7] https://www.k2view.com/what-is-synthetic-data-generation/
[8] https://www.neosync.dev/blog/referential-integrity