Method for Matrix Computation with Homomorphic Encryption

­Researchers have developed a general method based on homomorphic encryption for performing computations on sensitive data while guaranteeing privacy. More specifically, this secure matrix operation framework can protect data privacy or trade secret that have a sensitive nature, and have potential applications in Healthcare IT and Financial Section.

Background

Homomorphic encryption is a powerful method that allows computations to be performed on data without decrypting it and without access to a private key. Typically, sensitive information (such as genomic data) is stored in an encrypted format, but it must be decrypted before it can be analyzed.

Homomorphic encryption is the holy grail for the analysis of private health information, as it protects this highly sensitive information during the process of analysis, making cloud computing much more secure in the lifecycle of data outsourcing.  Recent advances in this field have made homomorphic encryption faster and reduced the amount of computing power needed, bringing us to the point where FHE can be practically implemented for the analysis of sensitive data in healthcare and financial applications.

Technology Highlights

• This framework can safeguard outsourcing sensitive private data in an untrusted 3^rd platform (cloud services or servers at another vendor);
• This framework can protect a developed model (e.g., predictive model for retinal disease diagnosis) to be securely evaluated without disclosing the actual implementation to the client;
• This novel framework can achieve high performance yet accurate ciphertext matrix operations compared to previous technologies.

Potential Users

• Healthcare institutions who want to exchange confidential information for research or collaboration with mitigated privacy risks;
• Machine Learning experts who build novel diagnosis models and want to protect their secret while providing services to end users;
• Genome data analysis companies who want to assemble large amount of rare disease patient data for privacy-preserving analysis.

Intellectual Property Status

• Issued U.S. Utility Patent 16/996,061
• Available for non-exclusive License

Stage of Development

Under development

Associated Publications

Secure Outsourced Matrix Computation and Application to Neural Networks

UTHealth Inventor

Xiaoqian Jiang, Ph.D.

• Professor of School of Biomedical Informatics (SBMI);
• Director of Center for Secure Artificial Intelligence for hEalthcare (SAFE).

Dr. Jiang’s research interest is to harmonize advanced machine learning and security/privacy technologies to develop privacy-preserving computational phenotyping models.

UTHealth Ref. No.: 2019-0005