This framework implements the data poisoning method found in the paper Adversarial Examples Make Strong Poisons

Overview

Adversarial poison generation and evaluation.

This framework implements the data poisoning method found in the paper Adversarial Examples Make Strong Poisons, authored by Liam Fowl, Micah Goldblum, Ping-yeh Chiang, Jonas Geiping, Wojtek Czaja, Tom Goldstein.

We use and adapt code from the publicly available Witches' Brew (Geiping et al.) github repository.

Dependencies:

  • PyTorch => 1.6.*
  • torchvision > 0.5.*

USAGE:

The cmd-line script anneal.py is responsible for generating poisons.

Other possible arguments for poison generation can be found under village/options.py. Many of these arguments do not apply to our implementation and are relics from the github repository which we adapted (see above).

Teaser

CIFAR-10 Example

Generation

To poison CIFAR-10 with our most powerful attack (class targeted), for a ResNet-18 with epsilon bound 8, use python anneal.py --net ResNet18 --recipe targeted --eps 8 --budget 1.0 --target_criterion reverse_xent --save poison_dataset_batched --poison_path /path/to/save/poisons --attackoptim PGD

  • Note 1: this will generate poisons according to a simple label permutation found in poison_generation/shop/forgemaster_targeted.py defined in the _label_map method. One can easily modify this to any permutation on the label space.

  • Note 2: this could take several hours depending on the GPU used. To decrease the time, use the flag --restarts 1. This will decrease the time required to craft the poisons, but also potentially decrease the potency of the poisons.

Generating poisons with untargeted attacks is more brittle, and the success of the generated poisons vary depending on the poison initialization much more than the targeted attacks. Because generating multiple sets of poisons can take a longer time, we have included an anonymous google drive link to one of our best untargeted dataset for CIFAR-10. This can be evaluated in the same way as the poisons generated with the above command, simply download the zip file from here and extract the data.

Evaluation

You can then evaluate the poisons you generated (saved in poisons) by running python poison_evaluation/main.py --load_path /path/to/your/saved/poisons --runs 1

Where --load_path specifies the path to the generated poisons, and --runs specifies how many runs to evaluate the poisons over. This will test on a ResNet-18, but this can be changed with the --net flag.

ImageNet

ImageNet poisons can be optimized in a similar way, although it requires much more time and resources to do so. If you would like to attempt this, you can use the included info.pkl file. This splits up the ImageNet dataset into subsets of 25k that can then be crafted one at a time (52 subsets in total). Each subset can take anywhere from 1-3 days to craft depending on your GPU resources. You also need >200gb of storage to store the generated dataset.

A command for crafting on one such subset is:

python anneal.py --recipe targeted --eps 8 --budget 1.0 --dataset ImageNet --pretrained --target_criterion reverse_xent --poison_partition 25000 --save poison_dataset_batched --poison_path /path/to/save/poisons --restarts 1 --resume /path/to/info.pkl --resume_idx 0 --attackoptim PGD

You can generate poisons for all of ImageNet by iterating through all the indices (0,1,2,...,51) of the ImageNet subsets.

  • Note: we are working to produce/run a deterministic seeded version of the above ImageNet generation and we will update the code appropriately.
CS506-Spring2022 - Code and Slides for Boston University CS 506

CS 506 - Computational Tools for Data Science Code, slides, and notes for Boston

Lance Galletti 17 May 06, 2022
Pytorch implementation of Masked Auto-Encoder

Masked Auto-Encoder (MAE) Pytorch implementation of Masked Auto-Encoder: Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick

Jiyuan 22 Dec 13, 2022
AOT (Associating Objects with Transformers) in PyTorch

An efficient modular implementation of Associating Objects with Transformers for Video Object Segmentation in PyTorch

162 Dec 14, 2022
OcclusionFusion: realtime dynamic 3D reconstruction based on single-view RGB-D

OcclusionFusion (CVPR'2022) Project Page | Paper | Video Overview This repository contains the code for the CVPR 2022 paper OcclusionFusion, where we

Wenbin Lin 193 Dec 15, 2022
MIMIC Code Repository: Code shared by the research community for the MIMIC-III database

MIMIC Code Repository The MIMIC Code Repository is intended to be a central hub for sharing, refining, and reusing code used for analysis of the MIMIC

MIT Laboratory for Computational Physiology 1.8k Dec 26, 2022
Semantic Segmentation with Pytorch-Lightning

This is a simple demo for performing semantic segmentation on the Kitti dataset using Pytorch-Lightning and optimizing the neural network by monitoring and comparing runs with Weights & Biases.

Boris Dayma 58 Nov 18, 2022
Suite of 500 procedurally-generated NLP tasks to study language model adaptability

TaskBench500 The TaskBench500 dataset and code for generating tasks. Data The TaskBench dataset is available under wget http://web.mit.edu/bzl/www/Tas

Belinda Li 20 May 17, 2022
Official code for "Decoupling Zero-Shot Semantic Segmentation"

Decoupling Zero-Shot Semantic Segmentation This is the official code for the arxiv. ZegFormer is the first framework that decouple the zero-shot seman

Jian Ding 108 Dec 30, 2022
SingleVC performs any-to-one VC, which is an important component of MediumVC project.

SingleVC performs any-to-one VC, which is an important component of MediumVC project. Here is the official implementation of the paper, MediumVC.

谷下雨 26 Dec 28, 2022
Deeper insights into graph convolutional networks for semi-supervised learning

deeper_insights_into_GCNs Deeper insights into graph convolutional networks for semi-supervised learning References data and utils.py come from Implem

Davidham3 17 Dec 16, 2022
Yolov5 deepsort inference,使用YOLOv5+Deepsort实现车辆行人追踪和计数,代码封装成一个Detector类,更容易嵌入到自己的项目中

使用YOLOv5+Deepsort实现车辆行人追踪和计数,代码封装成一个Detector类,更容易嵌入到自己的项目中。

813 Dec 31, 2022
This project hosts the code for implementing the ISAL algorithm for object detection and image classification

Influence Selection for Active Learning (ISAL) This project hosts the code for implementing the ISAL algorithm for object detection and image classifi

25 Sep 11, 2022
Easy to use and customizable SOTA Semantic Segmentation models with abundant datasets in PyTorch

Semantic Segmentation Easy to use and customizable SOTA Semantic Segmentation models with abundant datasets in PyTorch Features Applicable to followin

sithu3 530 Jan 05, 2023
P-Tuning v2: Prompt Tuning Can Be Comparable to Finetuning Universally Across Scales and Tasks

P-tuning v2 P-Tuning v2: Prompt Tuning Can Be Comparable to Finetuning Universally Across Scales and Tasks An optimized prompt tuning strategy for sma

THUDM 540 Dec 30, 2022
Shuffle Attention for MobileNetV3

SA-MobileNetV3 Shuffle Attention for MobileNetV3 Train Run the following command for train model on your own dataset: python train.py --dataset mnist

Sajjad Aemmi 36 Dec 28, 2022
The offcial repository for 'CharacterBERT and Self-Teaching for Improving the Robustness of Dense Retrievers on Queries with Typos', SIGIR2022

CharacterBERT-DR The offcial repository for CharacterBERT and Self-Teaching for Improving the Robustness of Dense Retrievers on Queries with Typos, Sh

ielab 11 Nov 15, 2022
Code for the ACL2021 paper "Lexicon Enhanced Chinese Sequence Labelling Using BERT Adapter"

Lexicon Enhanced Chinese Sequence Labeling Using BERT Adapter Code and checkpoints for the ACL2021 paper "Lexicon Enhanced Chinese Sequence Labelling

274 Dec 06, 2022
Fbone (Flask bone) is a Flask (Python microframework) starter/template/bootstrap/boilerplate application.

Fbone (Flask bone) is a Flask (Python microframework) starter/template/bootstrap/boilerplate application.

Wilson 1.7k Dec 30, 2022
Forecasting with Gradient Boosted Time Series Decomposition

ThymeBoost ThymeBoost combines time series decomposition with gradient boosting to provide a flexible mix-and-match time series framework for spicy fo

131 Jan 08, 2023
competitions-v2

Codabench (formerly Codalab Competitions v2) Installation $ cp .env_sample .env $ docker-compose up -d $ docker-compose exec django ./manage.py migrat

CodaLab 21 Dec 02, 2022