Date of Award

Fall 12-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical and Computer Engineering

Committee Director

Hongyi Wu

Committee Member

Chunsheng Xin

Committee Member

Jiang Li

Committee Member

Sachin Shetty

Abstract

Deep Learning (DL) has shown unrivalled performance in many applications such as image classification, speech recognition, anomalous detection, and business analytics. While end users and enterprises own enormous data, DL talents and computing power are mostly gathered in technology giants having cloud servers. Thus, data owners, i.e., the clients, are motivated to outsource their data, along with computationally-intensive tasks, to the server in order to leverage the server’s abundant computation resources and DL talents for developing cost-effective DL solutions. However, trust is required between the server and the client to finish the computation tasks (e.g., conducting inference for the newly-input data from the client, based on a well-trained model at the server) otherwise there could be the data breach (e.g., leaking data from the client or the proprietary model parameters from the server). Privacy-preserving DL takes data privacy into account where various data-encryption based techniques are adopted. However, the efficiency of linear and nonlinear computation for each DL layer remains a fundamental challenge in practice due to the intrinsic intractability and complexity of privacy-preserving primitives (e.g., Homomorphic Encryption (HE) and Garbled Circuits (GC)). As such, this dissertation targets deeply optimizing state-of-the-art frameworks as well as newly designing efficient modules by joint linear and nonlinear computation, with data encryption, to further boost the overall performance of privacy-preserving DL. Four contributions are made.

DOI

10.25777/6hgt-cv97

ISBN

9798762197410

Share

COinS