Black-Box Data Poisoning Attacks on Crowdsourcing (IJCAI23)

The GitHub repository for the paper "Black-Box Data Poisoning Attacks on Crowdsourcing" accepted by IJCAI 2023.

1 Introduction

This repository is built for the proposed method (SubPac). You can use anaconda's virtual environment to quickly reproduce the experiment results in this paper.

2 Usage

2.1 Requirement

• linux
• anaconda
• matlab (Exp 3 is needed)

2.2 Setup Virtual Environment

First, You can use the spec-list file to quickly build a conda virtual environment and install the required packages.

conda create  --name poisoning_attacks_on_crowdsourcing --file spec-list.txt
conda activate poisoning_attacks_on_crowdsourcing

2.3 Recurrence experiment

Senond, You can execute the python file to reproduce the experiment.

python exp1_proportion_of_malicious_workers.py
python exp2_worker_reliability.py
python exp3_transferability.py

After running the python file, you can find the text-based experimental results in the output folder. Not only that, you can also find the graphical experimental results in the plot folder.