adv-def-text-dist

Requirements:

• Python 3.6+
• TensorFlow 1.3+
• spacy
• transformers
• tensorflow_hub

Please refer to requirement.txt for detailed packages

Introduction

This is an implementation for the paper: Grey-box Adversarial Attack and Defence for Text. The paper is currently submitted to EMNLP'20.

Run training and test

1, Please download the Yelp review dataset from the official website link.

2, Download the GloVe embeddings and the counter-fitted embeddings using

./download.sh.

3, Run dataset preprecessing using

python yelp_preprocessing.py --data_dir YELP_DATASET_PATH --embed_file GLOVE_EMB_PATH

4, Train target models using the scripts

./train_cls.sh

5, Pre-train Auto-encoder for reconstruction using the scripts

./train_ae.sh.

6, Train adversarial attack/defence models using the scripts (multiple variants of our model are available and commented out in the script)

./train_adv.sh

7, Perform independent test for adversarial attack/defence using

./test_adv.sh