September 2022 Summaries

Differential Privacy is a mathematical framework designed to enable organizations to analyze datasets for patterns without compromising individual privacy. This approach addresses the limitations of traditional anonymization techniques, which often fail to prevent data re-identification. By introducing controlled randomness or "noise," Differential Privacy ensures that the output of data analysis remains consistent whether or not any individual's data is included, thus providing quantifiable privacy guarantees. There are two main deployment types: Central Differential Privacy (CDP), where a trusted curator applies privacy mechanisms, and Local Differential Privacy (LDP), where individuals modify their own data. While LDP is more secure due to decentralized data handling, it is less efficient compared to CDP. Noteworthy applications include Apple's use of Differential Privacy in iOS and macOS, and Google's implementation of privacy-preserving tools in Chrome. However, Differential Privacy is not a cure-all; it is more effective with large datasets and may require integration with technologies like full homomorphic encryption to maintain privacy in collaborative environments. Despite its advantages, such as compliance with privacy regulations and resistance to adversarial attacks, Differential Privacy has limitations, particularly in small datasets and repeated data queries, which can lead to privacy leaks.

Duality Technologies, a leader in privacy-preserving data collaboration, announced that Google has integrated its Fully Homomorphic Encryption (FHE) Transpiler with the Duality-led OpenFHE library to enhance the accessibility and adoption of FHE technologies among developers. This collaboration aims to democratize the use of advanced cryptographic methods by allowing developers to convert high-level code into formats that operate on encrypted data without requiring in-depth cryptographic knowledge. OpenFHE, based on post-quantum open-source lattice cryptography, provides a modular and portable framework designed for ease of use and integration. Google's Transpiler simplifies the process of implementing FHE, making it more accessible for software designers and developers by bridging the knowledge gap associated with cryptographic expertise. The partnership is expected to further the application of FHE in practical fields like machine learning, aligning with Duality's strategic goals of enhancing data privacy and regulatory compliance while enabling advanced data insights.

The interview between Prof. Michael Jordan and Prof. Shafi Goldwasser delves into the complexities of ethical AI, focusing on how human values such as privacy, fairness, transparency, and trust must be integrated into AI systems to ensure ethical decision-making. They explore the intersection of economics, computer science, and statistics in addressing real-world problems, emphasizing the importance of incorporating social sciences and law in the responsible deployment of AI technologies. The discussion highlights the necessity of public education on AI terminology and its implications, as well as strategies to build trust in machine-learning systems through audit, recourse, and benchmarking. Prof. Jordan stresses the importance of developing systematic approaches to the interaction between technology and humans, underscoring the potential benefits of ethical AI not only for society but also for business.

Homomorphic Encryption (HE) is a critical technology for enabling computations on encrypted data, allowing data to remain private and secure even when processed on untrusted systems. OpenFHE is an advanced open-source library that facilitates the use of HE, and Duality employs it to support privacy-preserving data collaboration. HE schemes, such as BGV, BFV, and CKKS, allow vectorized operations on encrypted data, whereas TFHE and FHEW, collectively known as BinFHE, focus on Boolean computations. Despite the advantages of HE, it imposes computational and storage overheads and lacks certain operations, which researchers address using approximation techniques. The Google Transpiler, an open-source tool, translates C++ code into digital circuits compatible with OpenFHE, enabling complex encrypted operations. This innovation, alongside OpenFHE, supports applications like shortest path calculations by converting data structures into encrypted forms for secure computation. Future developments aim to enhance OpenFHE's capabilities, including interoperability between different HE schemes and expanding support for additional data types and operations.

Thousands of attendees gathered at the MIT CDOIQ 2022 Symposium to discuss challenges and solutions in the data space, with sessions focusing on sharing sensitive data, CDO tenure, and digital transformation within the Department of Defense. Key discussions included the strategic yet challenging nature of sharing sensitive data, with speakers advocating for privacy-preserving technologies like differential privacy and federated learning. Insights were shared on the short average tenure of Chief Data Officers (CDOs) compared to other C-level executives, highlighting the need for CDOs to be treated as equal partners and prioritize data accessibility to drive organizational success. The event also explored the role of digital transformation in the US Government, emphasizing the use of cutting-edge technologies to modernize data acquisition strategies. Participants expressed optimism about overcoming privacy challenges to extend CDO tenure by leveraging solutions like Duality’s secure data collaboration platform.

The Gartner Data and Analytics Summit 2022 in Orlando highlighted various trends and challenges in the data industry, with a significant focus on synthetic data and its role in training models without exposing real data to risks. However, the event notably lacked discussions on data collaboration, a crucial element for overcoming data silos and enabling secure, privacy-preserving data sharing. The misconception that synthetic data alone can address secure data collaboration needs suggests a gap in understanding Privacy Enhancing Technologies (PETs), which could potentially solve these issues. The Duality privacy-preserving collaboration platform was presented as a solution for applying models to sensitive data without exposing it, facilitating cross-departmental and cross-border collaboration. The platform's ability to securely link multiple real data sets using post-quantum encryption offers enterprises a chance to unify data previously constrained by privacy limitations. This approach not only protects sensitive data but also safeguards proprietary models, enhancing monetization opportunities. Despite the promise of PETs in breaking down data silos, there remains skepticism about their practicality, likening them to science fiction. However, the Duality Platform demonstrates the potential to unlock valuable insights from sensitive data while maintaining privacy.

Duality has been recognized in the 2022 Gartner Hype Cycle for Data Security as a Sample Vendor for Homomorphic Encryption, a technology that allows data analytics on encrypted data without exposing it, thereby enhancing privacy and data processing. This advancement enables secure data collaboration among businesses and third parties, allowing them to perform tasks like sharing anti-money laundering data without compromising privacy. As organizations strive to utilize sensitive data for better decision-making, homomorphic encryption is highlighted by Gartner analysts as a transformational data security technology, offering prospects for data sharing and pooling without revealing secrets. Duality supports enterprises in safely extracting business value from sensitive data through secure collaborations, addressing the growing need for data privacy solutions amidst increasing data-driven business objectives.