The goal of this project is to understand (and optimize) the system-level bottlenecks of private inference protocols under varying workloads. This repo contains the codes for our paper. The high-level directories are:
garbled_circuits
: contains the raw data for benchmarking ReLU Garbling and Evaluation on the embedded device and the serverlayer_parallel_HE
: contains our code and the raw data for enabling layer-parallel homomorophic evaluation of linear layerssimulator
: our PI simulator used to generate plots and results from the paperartifact
: contains scripts to generate key figures in the paper
Please refer to the READMEs within each subdirectory to learn how we collected data and how to run the simulator.
First, clone this repo:
git clone https://github.com/kvgarimella/characterizing-private-inference.git
Then, change directories and install the Python dependencies:
cd characterizing-private-inference
pip install -r requirements.txt
As a simple test, navigate to the simulator/experiments
sub-directory and run:
cd simulator/experiments
python simulate_server_garbler.py
This will run a single simulation of the Server-Garbler protocol and store the results in a directory called tmp
.
Please refer to the README within the artifact
directory.
If you find our work useful, please use the following citation:
@inproceedings{10.1145/3582016.3582065,
author = {Garimella, Karthik and Ghodsi, Zahra and Jha, Nandan Kumar and Garg, Siddharth and Reagen, Brandon},
title = {Characterizing and Optimizing End-to-End Systems for Private Inference},
year = {2023},
publisher = {Association for Computing Machinery}, address = {New York, NY, USA},
url = {https://doi.org/10.1145/3582016.3582065}, doi = {10.1145/3582016.3582065},
booktitle = {Proceedings of the 28th ACM International Conference on
Architectural Support for Programming Languages and Operating Systems, Volume 3},
series = {ASPLOS 2023}}
- Karthik Garimella - Python, Rust, C++
- Zahra Ghodsi - Rust, C++