Abstract

Homomorphic encryption (HE) is a promising cryptographic technique that supports computations on encrypted data without requiring decryption first. This ability allows sensitive data, such as genomic, financial, or location data, to be outsourced for evaluation in a resourceful third-party such as the cloud without compromising data privacy. Basic homomorphic primitives support addition and multiplication on ciphertexts. These primitives can be utilized to represent essential computations, such as logic gates, which subsequently can support more complex functions. We propose the construction of efficient cryptographic protocols as building blocks (e.g., equality, comparison, and counting) that are commonly used in data analytics and machine learning. We explore the use of these building blocks in two privacy-preserving applications. One application leverages our secure prefix matching algorithm, which builds on top of the equality operation, to process geospatial queries on encrypted locations. The other applies our secure comparison protocol to perform conditional branching in private evaluation of decision trees.

There are many outsourced computations that require joint evaluation on private data owned by multiple parties. For example, Genome-Wide Association Study (GWAS) is becoming feasible because of the recent advances of genome sequencing technology. Due to the sensitivity of genomic data, this data is encrypted using different keys possessed by different data owners. Computing on ciphertexts encrypted with multiple keys is a non-trivial task. Current solutions often require a joint key setup before any computation such as in threshold HE or incur large ciphertext size (at best, grows linearly in the number of involved keys) such as in multi-key HE. We propose a hybrid approach that combines the advantages of threshold and multi-key HE to support computations on ciphertexts encrypted with different keys while vastly reducing ciphertext size.

Moreover, we propose the SparkFHE framework to support large-scale secure data analytics in the Cloud. SparkFHE integrates Apache Spark with Fully HE to support secure distributed data analytics and machine learning and make two novel contributions: (1) enabling Spark to perform efficient computation on large datasets while preserving user privacy, and (2) accelerating intensive homomorphic computation through parallelization of tasks across clusters of computing nodes. To our best knowledge, SparkFHE is the first addressing these two needs simultaneously.

Library of Congress Subject Headings

Cloud computing--Security measures; Homomorphisms (Mathematics); Data encryption (Computer science)

Publication Date

12-7-2020

Document Type

Dissertation

Student Type

Graduate

Degree Name

Computing and Information Sciences (Ph.D.)

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Peizhao Hu

Advisor/Committee Member

Stanisław Radziszowski

Advisor/Committee Member

Matthew Wright

Campus

RIT – Main Campus

Plan Codes

COMPIS-PHD

Share

COinS