January 2023 Summaries

The healthcare industry faces significant challenges in data collaboration due to stringent regulations on Personal Health Information (PHI), which hinder data sharing between medical centers and within institutions, impacting clinical trials, treatment development, and regimen effectiveness assessments. Technologies like Duality offer solutions by enabling the analysis of encrypted data without compromising privacy, facilitating potential cross-departmental data analysis, genomic data integration, and real-world evidence collection. These advancements could significantly benefit areas such as rare disease research, where limited data and economic incentives currently impede progress; by improving data collaboration, the quality of care and treatment options for rare diseases could be enhanced, potentially altering the landscape of what is considered "rare."

Organizations increasingly face the challenge of protecting sensitive data while collaborating, particularly when the data is in use and vulnerable to cyberattacks. Fully Homomorphic Encryption (FHE) emerges as a promising solution by allowing computations on encrypted data without decryption, thus preserving privacy, especially in machine learning applications. However, implementing FHE for machine learning is computationally intensive and requires substantial resources, particularly due to the need for bootstrapping. Hardware acceleration, using platforms like FPGA, GPU, and ASIC, is seen as a viable solution to improve efficiency and make FHE practical. The OpenFHE Library supports this by providing a community-driven, open-source cryptographic software framework, which includes a unique Hardware Abstraction Layer (HAL) to integrate different hardware acceleration technologies. This development opens opportunities for hardware providers to advance FHE applications, contributing to the growing trend of privacy-preserving machine learning.

Bootstrapping, a complex and computationally intensive component of Fully Homomorphic Encryption (FHE), is central to enabling the evaluation of arbitrary computations on encrypted data without accessing the secret key. The concept, first proposed by Craig Gentry in 2008, involves homomorphically evaluating the decryption procedure to refresh noisy ciphertexts, allowing further computations. This document aims to clarify bootstrapping's intricacies, correct common misconceptions, and compare various methods, such as those used in CKKS, DM/CGGI, and BGV/BFV schemes, highlighting their respective trade-offs and performance. Each method's suitability varies depending on the application, with CKKS being optimal for real-number computations, DM/CGGI for arbitrary functions over small integers, and BGV/BFV for large vectors of small integers. The document also discusses practical guidelines and the potential for hardware acceleration to improve efficiency, offering recommendations for deploying FHE in different contexts. The OpenFHE library incorporates bootstrapping for several schemes, with ongoing developments to enhance its capabilities.

Duality Technologies, a leader in privacy-preserving data collaboration, has been selected for Phase II of the DARPA Data Protection in Virtual Environments (DPRIVE) program, building on its success in Phase I. The company is integrating new capabilities for accelerated privacy-protected machine learning into its secure data collaboration platform, which aims to facilitate data science and machine learning on sensitive data while breaking down organizational data silos. The Duality TREBUCHET team is leveraging the Duality-powered OpenFHE library for Fully Homomorphic Encryption (FHE) to reduce computational burdens and accelerate FHE calculations. Their goal is to make machine learning on FHE-protected ciphertext as fast as on unencrypted data by designing a hardware accelerator integrated with advanced FHE software. This research, partially funded by the U.S. Government, supports applications for the Department of Defense and commercial sectors, with the team's work marking significant strides in making privacy technology practical and efficient.

Creating a data collaboration strategy is essential for organizations to fully utilize their data while addressing privacy concerns, and the Duality Platform offers a solution by enabling secure and efficient data collaboration. The platform supports three main models of data analysis: single dataset analytics, union dataset analytics (horizontal data analysis), and join dataset analytics (vertical data analysis), each offering unique benefits for different collaboration needs. By integrating advanced cryptography and data science techniques, Duality allows organizations to perform secure data sharing and analysis, including federated learning, while maintaining data privacy and integrity. The platform also provides tools for schema collaboration, data preprocessing, and entity resolution, facilitating seamless and secure data collaboration that unlocks new insights without compromising sensitive information.

Over the past decade, the integration of data and advanced technologies has been increasingly recognized as crucial for digital transformation in healthcare, particularly through the use of Privacy Enhancing Technologies (PETs) that allow for secure data collaboration. The application of machine learning (ML) in healthcare is exemplified by companies like Bayesian Health, which uses ML for early detection of conditions like sepsis, potentially reducing mortality rates. The transition to digital health data has faced challenges, including defining success benchmarks, overcoming industry hesitation, and updating outdated infrastructure to support real-time analytics. Duality Technologies offers solutions that enable secure and compliant data collaboration across healthcare institutions, utilizing federated analytics and cryptographic tools. As the healthcare industry continues to embrace digital transformation, there is a growing emphasis on educating the public about the benefits and rigorous validation of advanced algorithms, with the potential to significantly reduce healthcare costs and mortality rates within the next decade.