Skip to content
/ E3 Public

Code for the paper 'Efficient and Effective Model Extraction' accepted at IEEE ICASSP 2025.

Notifications You must be signed in to change notification settings

GradOpt/E3

Repository files navigation

Efficient and Effective Model Extraction

Code for the paper 'Efficient and Effective Model Extraction' accepted at IEEE ICASSP 2025.


Usage

The core implementation is now available. Experimental parameters on CIFAR-10 are listed in the corresponding .py files.

To run E3:

  1. First, execute query_selection.py to generate the query set from the candidate out-of-distribution (OOD) set.
  2. Then, run model_extraction.py to perform the extraction attack.
  3. Finally, if desired, execute distribution_alignment.py to apply test-time distribution alignment.

About

Code for the paper 'Efficient and Effective Model Extraction' accepted at IEEE ICASSP 2025.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages