Enabling Homomorphically Encrypted Inference for Large DNN Models

TL;DR

This paper demonstrates that using hybrid DRAM and persistent memory systems, specifically Intel Optane PMem, makes homomorphically encrypted inference of large DNNs feasible and efficient, overcoming previous resource limitations.

Contribution

It introduces a novel approach leveraging hybrid memory systems and HE-Transformer nGraph to enable large DNN inference with homomorphic encryption, which was previously impractical due to resource constraints.

Findings

Homomorphic encryption inference benefits from hybrid memory access patterns.

Large models like MobileNetV2 and ResNet-50 can be run efficiently with HE on new hardware.

Analysis shows improved performance with specific hardware and software configurations.

Abstract

The proliferation of machine learning services in the last few years has raised data privacy concerns. Homomorphic encryption (HE) enables inference using encrypted data but it incurs 100x-10,000x memory and runtime overheads. Secure deep neural network (DNN) inference using HE is currently limited by computing and memory resources, with frameworks requiring hundreds of gigabytes of DRAM to evaluate small models. To overcome these limitations, in this paper we explore the feasibility of leveraging hybrid memory systems comprised of DRAM and persistent memory. In particular, we explore the recently-released Intel Optane PMem technology and the Intel HE-Transformer nGraph to run large neural networks such as MobileNetV2 (in its largest variant) and ResNet-50 for the first time in the literature. We present an in-depth analysis of the efficiency of the executions with different hardware and software configurations. Our results conclude that DNN inference using HE incurs on friendly access patterns for this memory configuration, yielding efficient executions.