Adversarial Attacks on Question Answering models
Code to reproduce results in the paper
Mudrakarta, Pramod Kaushik, Ankur Taly, Mukund Sundararajan, and Kedar Dhamdhere. "Did the Model Understand the Question?." ACL 2018.
@article{mudrakarta2018did,
title={Did the Model Understand the Question?},
author={Mudrakarta, Pramod Kaushik and Taly, Ankur and Sundararajan, Mukund and Dhamdhere, Kedar},
journal={arXiv preprint arXiv:1805.05492},
year={2018}
}
Setup
Clone the repository using
git clone https://github.com/pramodkaushik/acl18_results.git --recursive
Code for experiments
Attacks on Neural Programmer are present in the folder np_analysis
, and those on visual question answering in visual_qa_analysis
. Code for computing attributions via Integrated Gradients and to reproduce experiments are in Jupyter notebooks in both these directories.
Contact
Pramod Kaushik Mudrakarta