ObfuNAS

This is the offcial implementation of ObfuNAS (ICCAD 2022) on NASbench-101 dataset.

Description

This framework is proposed to defend against model architecture extraction by obfuscating the victim model architracture as a different one, while considering the accuracy the obfuscated model may achieve. ObfuNAS converts the DNN architecture obfuscation into a neural architecture search (NAS) problem. Using a combination of function-preserving obfuscation strategies, it ensures that the obfuscated DNN architecture can only achieve lower accuracy than the victim.

Getting Started

Requirement

Python 3.7
torch 1.10
nasbench library:

git clone https://github.com/google-research/nasbench

Download

Dataset (provided by nasbench): https://storage.googleapis.com/nasbench/nasbench_only108.tfrecord
More details are in the nasbench repo: https://github.com/google-research/nasbench/tree/master

Executing program

Change the victim spec in main.py/ evaluate.py and run

python main.py