Methods
Providing high-quality synthetic data that balances privacy and utility is at the heart of the Octopize methodology. Our goal is to make the generation of such high-quality data as simple and accessible as possible for every user—no matter their technical background. Because data comes in many forms, a single anonymization method simply isn't enough. That's why we've developed a set of specialized techniques—each tailored to a specific data structure—to ensure adaptability across all situations.
We distinguish three main data structures, each requiring a dedicated framework to use our algorithms.
Note: All methods developed regarding a specific data structure are compatible with each other. We can for instance apply both Single Tabular and Time Series anonymization within a Multi Table anonymization context.
1. Single Tabular Data
For datasets where each row corresponds to a unique individual, the single-table framework is the most suitable. Simple tabular data was the first use case we addressed historically. Over time, we have developed multiple methodologies to handle this structure, offering flexibility depending on the context and privacy needs.
- The standard Avatar method, remains our go-to approach. this method efficiently preserves statistical properties while ensuring privacy at the individual level.
🔗 Learn more about Single Tabular Data anonymization - We also implemented AvatarFastDP, a variant of the Avatar method that generates synthetic data with differential privacy mechanisms.
🔗 Learn more about Differential Privacy methods
2. Multi-Table (Relational) Data
When information is spread across multiple linked tables (e.g., patient records and medical visits), a multi-table framework is required. This method maintains consistency across relationships while anonymizing both individual tables and their interconnections.
🔗 Learn more about Multi-Table Data anonymization
When the tables live in a database, Avatar can read them there and write the anonymized copy into another database, keys and joins included.
🔗 Learn more about Database anonymization
3. Time Series Data
For datasets where values evolve over time (e.g., sensor readings, medical signals), a time-series-adapted method should be applied. This framework ensures that trends, patterns, and temporal dependencies are preserved while anonymizing sensitive information.
🔗 Learn more about Time Series Data anonymization
The Avatar methodology is designed to generate synthetic data that maintain the structure, granularity, and statistical relevance of the original data while ensuring compliance with privacy regulations. We have built multiple frameworks to help you achieve both data protection and analytical integrity for any data structure.