ranran0523/SPECNN

code repo for paper accepted in ICML 2023

SPENCNN

This is the implementation for ICML2023 paper:

SpENCNN: Orchestrating Encoding and Sparsity for Fast Homomorphically Encrypted Neural Network Inference (https://proceedings.mlr.press/v202/ran23b/ran23b.pdf)

Installation

pip install -r requirements.txt

HE-group convolution

By running the 3 ipynb file to perform group convolution replaced CNNs training.

HE-inference Test

To use a multi-thread simulation environment on testing inference latency, we fork the code from HEAR(https://github.com/K-miran/HEAR).

Acknowledgement

Code forked and modified from https://github.com/wanglouis49/pytorch-weights_pruning