Learnability-Lock

This is the official code for our paper Learnability Lock: Authorized Learnability Control Through Adversarial Invertible Transformations, accepted at ICLR 2022.

Refer to QuickStart notebook to generate unlearnable examples on CIFAR10.

Prerequisites

• Python 3.6
• Pytorch
• Numpy

Usage

For i-Resnet block based lock, initialize using:

// customize the params following QuickStart
params = {'in_shape':32, 
               'n_channel':3, 
               'n_class':10,
               'mid_planes':32}
lock = iResLock(lock_params = params, epsilon=<>)

Linear transformation lock:

lock = LinearLock(lock_params = params, epsilon=<>)