Sphynx: ReLU-Efficient Network Design for Private Inference

TL;DR

Sphynx introduces a ReLU-efficient neural network design optimized for private inference, significantly reducing latency and outperforming existing methods on CIFAR-100, Tiny-ImageNet, and ImageNet datasets.

Contribution

The paper proposes a micro-search based method to design ReLU-efficient networks tailored for private inference, achieving state-of-the-art performance and scalability.

Findings

Outperforms all existing private inference methods on CIFAR-100

Enables cryptographically private inference on large-scale datasets

Reduces latency caused by non-linear operations in neural networks

Abstract

The emergence of deep learning has been accompanied by privacy concerns surrounding users' data and service providers' models. We focus on private inference (PI), where the goal is to perform inference on a user's data sample using a service provider's model. Existing PI methods for deep networks enable cryptographically secure inference with little drop in functionality; however, they incur severe latency costs, primarily caused by non-linear network operations (such as ReLUs). This paper presents Sphynx, a ReLU-efficient network design method based on micro-search strategies for convolutional cell design. Sphynx achieves Pareto dominance over all existing private inference methods on CIFAR-100. We also design large-scale networks that support cryptographically private inference on Tiny-ImageNet and ImageNet.