5th Year Thesis Presentation - Trevor Leong

— 5:00pm

Location:
In Person - Gates Hillman 9115

Speaker:
TREVOR LEONG, Master's Student, Computer Science Department, Carnegie Mellon University

Improvements to the evaluation of ML models in FHE

FHE (Fully Homomorphic encryption) is a class of encryption schemes that enables computation over ciphertexts that correspond to operations on the encrypted plaintexts. This useful property allows clients to outsource computation to servers without revealing what their inputs were. A notable application of FHE is in private Machine Learning as a Service (MLaaS), where clients can submit data to a server-hosted machine learning model and receive processed results, all while preserving data confidentiality.

However, the practical implementation of FHE in evaluating machine learning models remains challenging. A significant hurdle is the restricted set of operations permissible under FHE. This is compounded by the fact that each FHE operation incurs a significant performance overhead when compared to its plaintext counterpart. As a result, computation of nonlinear functions like Softmax requires complex polynomial approximations and even basic operations such as Matrix multiplication require a significant amount of time to evaluate.

This thesis seeks to mitigate these performance constraints associated with FHE in machine learning model evaluation. Initially, I introduce a modification to the HELR algorithm, enhancing its efficiency when deployed on a hardware accelerator, achieving an 8x reduction in latency. Then, I propose a novel application of a softmax approximation for evaluation in FHE that leads to a 4x reduction in latency. Lastly, I describe a procedure for evaluating the embedding layer on the server without the client learning the model’s embedding matrix at minimal computation overhead.

Thesis Committee:

Wenting Zheng (Advisor)
Lujo Bauer

In Person and Zoom Participation.  See announcement.