[ICLR2021] Unlearnable Examples: Making Personal Data Unexploitable

Last update: Dec 07, 2022

Related tags

Overview

Unlearnable Examples

Code for ICLR2021 Spotlight Paper "Unlearnable Examples: Making Personal Data Unexploitable " by Hanxun Huang, Xingjun Ma, Sarah Monazam Erfani, James Bailey, Yisen Wang.

Quick Start

Use the QuickStart.ipynb notebook for a quick start.

In the notebook, you can find the minimal implementation for generating sample-wise unlearnable examples on CIFAR-10. Please remove mlconfig from models/__init__.py if you are only using the notebook and copy-paste the model to the notebook.

Experiments in the paper.

Check scripts folder for *.sh for each corresponding experiments.

Sample-wise noise for unlearnable example on CIFAR-10

Generate noise for unlearnable examples

python3 perturbation.py --config_path             configs/cifar10                \
                        --exp_name                path/to/your/experiment/folder \
                        --version                 resnet18                       \
                        --train_data_type         CIFAR-10                       \
                        --noise_shape             50000 3 32 32                  \
                        --epsilon                 8                              \
                        --num_steps               20                             \
                        --step_size               0.8                            \
                        --attack_type             min-min                        \
                        --perturb_type            samplewise                      \
                        --universal_stop_error    0.01

Train on unlearnable examples and eval on clean test

python3 -u main.py    --version                 resnet18                       \
                      --exp_name                path/to/your/experiment/folder \
                      --config_path             configs/cifar10                \
                      --train_data_type         PoisonCIFAR10                  \
                      --poison_rate             1.0                            \
                      --perturb_type            samplewise                      \
                      --perturb_tensor_filepath path/to/your/experiment/folder/perturbation.pt \
                      --train

Class-wise noise for unlearnable example on CIFAR-10

Generate noise for unlearnable examples

python3 perturbation.py --config_path             configs/cifar10                \
                        --exp_name                path/to/your/experiment/folder \
                        --version                 resnet18                       \
                        --train_data_type         CIFAR-10                       \
                        --noise_shape             10 3 32 32                     \
                        --epsilon                 8                              \
                        --num_steps               1                              \
                        --step_size               0.8                            \
                        --attack_type             min-min                        \
                        --perturb_type            classwise                      \
                        --universal_train_target  'train_subset'                 \
                        --universal_stop_error    0.1                            \
                        --use_subset

Train on unlearnable examples and eval on clean test

python3 -u main.py    --version                 resnet18                       \
                      --exp_name                path/to/your/experiment/folder \
                      --config_path             configs/cifar10                \
                      --train_data_type         PoisonCIFAR10                  \
                      --poison_rate             1.0                            \
                      --perturb_type            classwise                      \
                      --perturb_tensor_filepath path/to/your/experiment/folder/perturbation.pt \
                      --train

Cite Our Work

@inproceedings{huang2021unlearnable,
    title={Unlearnable Examples: Making Personal Data Unexploitable},
    author={Hanxun Huang and Xingjun Ma and Sarah Monazam Erfani and James Bailey and Yisen Wang},
    booktitle={ICLR},
    year={2021}
}

[ICLR2021] Unlearnable Examples: Making Personal Data Unexploitable

Related tags

Overview

Unlearnable Examples

Quick Start

Use the QuickStart.ipynb notebook for a quick start.

Experiments in the paper.

Sample-wise noise for unlearnable example on CIFAR-10

Generate noise for unlearnable examples

Train on unlearnable examples and eval on clean test

Class-wise noise for unlearnable example on CIFAR-10

Generate noise for unlearnable examples

Train on unlearnable examples and eval on clean test

Cite Our Work

Owner

Hanxun Huang

PyTorch code for ICLR 2021 paper Unbiased Teacher for Semi-Supervised Object Detection

Artificial Neural network regression model to predict the energy output in a combined cycle power plant.

KITTI-360 Annotation Tool is a framework that developed based on python(cherrypy + jinja2 + sqlite3) as the server end and javascript + WebGL as the front end.

Fully Convolutional Refined Auto Encoding Generative Adversarial Networks for 3D Multi Object Scenes

MegEngine implementation of YOLOX

Council-GAN - Implementation for our paper Breaking the Cycle - Colleagues are all you need (CVPR 2020)

A transformer model to predict pathogenic mutations

Learning to Stylize Novel Views

A simplistic and efficient pure-python neural network library from Phys Whiz with CPU and GPU support.

Understanding and Overcoming the Challenges of Efficient Transformer Quantization

pip install python-office

DNA sequence classification by Deep Neural Network

Unofficial implement with paper SpeakerGAN: Speaker identification with conditional generative adversarial network

The AWS Certified SysOps Administrator

Automatic Number Plate Recognition using Contours and Convolution Neural Networks (CNN)

Keywords : Streamlit, BertTokenizer, BertForMaskedLM, Pytorch

Official PyTorch implementation of MX-Font (Multiple Heads are Better than One: Few-shot Font Generation with Multiple Localized Experts)

A Runtime method overload decorator which should behave like a compiled language

Modified fork of Xuebin Qin's U-2-Net Repository. Used for demonstration purposes.

Rethinking Transformer-based Set Prediction for Object Detection