Frequently Asked Questions
TL;DR: We allow analytics and ML to be applied on sensitive data, providing mathematically guaranteed private outputs by introducing randomization to the computation.
Differential privacy (DP) is a set of systems and practices that help keep the data of individuals safe and private. Differential Privacy offers the strongest possible privacy protection available today, with a mathematical guarantee to back up each algorithm. Differential privacy is achieved by introducing statistical noise. The noise is significant enough to protect the privacy of any individual in the data, but small enough that it will not impact the accuracy of analytics and machine learning methods applied on the data.
PVML offers proprietary Differential Privacy technology to exract useful insights and train AI models using datasets containing sensitive information. Our algorithms are performed on the analysis itself, on-the-fly, so that the outputs are privacy-preserving and can be safely used or shared by the user or third-party.
TL;DR: As opposed to Homomorphic Encryption, Differential Privacy has no overhead in computation and memory cost, and it also guarantees privacy at the output level, preventing reverse engineering and attribute inference attacks.
Homomorphic Encryption allows computation directly on encrypted data, however – it isn’t efficient. Because Homomorphic Encryption comes with a large performance overhead, computations that are already costly to do on unencrypted data probably aren’t feasible on encrypted data. Moreover, although the data is unreadable, the computations performed on it remain the same, including the outputs. When outputs are returned in perfect accuracy, the privacy of individuals in the data cannot be guaranteed, and the dataset remains vulnerable to re-identification attacks where sensitive raw data may be extracted in reverse engineering and attribute inference attacks.
TL;DR: PVML prioritizes applicable algorithmic capabilities, beyond what science can currently provide in the field of Differential Privacy.
PVML incorporates beyond state-of-the-art research objectives along with software engineering and applied machine learning in order to provide the most efficient Differential Privacy algorithms that produce privacy-preserving results with higher accuracy than existing Differential Privacy solutions. Applicability is our first priority, ensuring that our Differential Privacy algorithms can be seamlessly integrated into a wide range of applications and systems, and without changing the methods, tools or languages you use to interact with data. Whether you are in healthcare, finance, telecommunications, or any other industry, our cutting-edge solutions are designed to safeguard sensitive information while maintaining the utility and integrity of your data. Our commitment to applicability extends to easy deployment, scalability, and adaptability, allowing organizations of all sizes to benefit from state-of-the-art privacy protection without compromising performance.
TL;DR: PVML has been verified by legal and technological experts in the privacy field.
The legislation mandates companies to design their products and processes with privacy in mind, meaning that a company is responsible for ensuring and maintaining the privacy of the personal data it handles. We work alongside a legal team and various security and privacy experts who provide guidance and validation throughout our development process, thereby ensuring that our Differential Privacy algorithms and overall approach maintain individuals’ privacy in accordance with various privacy regulations. Furthermore, we undergo rigorous external audits to ensure that our solution adheres to the highest standards of privacy and security and is SOC2 compliant.
TL;DR: Yes, anonymization is an outdated technique that leaves expensive data value on the table and fails to guarantee privacy, especially in the current age of AI.
Yes! Even when removing personally identifiable information (PIIs), the resulting records often include unique combinations of variables and features that might be linked to other publicly available information in order to re-identify specific people or leak sensitive information. In practice, as long as useful information about individuals is included in the data, it is vulnerable to re-identification attacks (and therefore, not anonymous).
Moreover, as we transition into an era where data is not only accessed by people but increasingly by advanced AI systems, the risks escalate. AI, being smarter, faster, and exposed to a wealth of information, introduces new challenges to traditional anonymization methods. These intelligent systems can perform intricate attribute inferencing, extracting nuanced insights and patterns that may not be readily apparent to human users. This capability, if exploited by human users, poses significant risks of intentional misuse. Moreover, there’s a potential for unintentional mistakes by AI, leading to inadvertent exposure of sensitive information, further amplifying the challenges in safeguarding data integrity and privacy.
Therefore, the evolving landscape of technology requires a comprehensive approach to anonymization to safeguard against risks posed by both human and AI access. PVML’s data protection technology is grounded in mathematics and engineered for the age of AI, ensuring heightened protection against data vulnerabilities and privacy breaches regardless of whether data is accessed by human users, applications, or AI models.
Your sensitive data stays wherever it is located (on-premise / on-cloud) and our platform does not require any duplication or modification of the data.