kalmzzz / Explain2Attack

Implementation of the paper: Explain2Attack: Text Adversarial Attacks via Cross-Domain Interpretability (ICPR 2020)

Explain2Attack

Implementation for the paper: "Explain2Attack: Text Adversarial Attacks via Cross-Domain Interpretability", In 25th International Conference on Pattern Recognition (ICPR 2020).

Prerequisites:

• Pytorch >= 0.4
• Tensorflow >= 1.0
• Numpy
• Python >= 3.6

How to use

• Run the following code to install the esim package:

cd ESIM
python setup.py install
cd ..

• Please follow download_instructions.txt inside subfolders to download missing data files.
• (Optional) Run the following code to pre-compute the cosine similarity scores between word pairs based on the counter-fitting word embeddings.

python comp_cos_sim_mat.py [PATH_TO_COUNTER_FITTING_WORD_EMBEDDINGS]

• Run the following code to generate the adversaries for text classification:

python attack_classification.py

Example to reproduce IMDB results:

python attack_classification.py --l2x_train_data_size 25000 --l2x_train_data_path data/amazon_movies_20K/train_tok.csv --l2x_test_data_size 25000 --l2x_test_data_path data/amazon_movies_20K/test_tok.csv --l2x_max_seq_length 256 --l2x_k_words 20 --l2x_bert_tokenize no --attack_data_size 25000 --data_format_label_fst yes --attack_data_path data/imdb/test_tok.csv --word_rank_method l2x --target_model wordLSTM --target_model_path classifiers/lstm/imdb_trim_len_256_0 --word_embeddings_path data/embeddings/glove.6B/glove.6B.200d.txt --attack_max_seq_length 256 --batch_size 32 --counter_fitting_embeddings_path counter-fitting/word_vectors/counter-fitted-vectors.txt --counter_fitting_cos_sim_path ./cos_sim_counter_fitting.npy --USE_cache_path ./tf_cache --l2x_remove_stop_words yes --rank_only no --l2x_rseed 10086 --sim_score_threshold 0.7 --outdir_postfix scores_len_max_attack_len

This code is based on the source code for the paper: Jin, Di, et al. "Is BERT Really Robust? Natural Language Attack on Text Classification and Entailment." arXiv preprint arXiv:1907.11932 (2019).

To cite this work, please use

@misc{hossam2020explain2attack,
      title={{Explain2Attack: Text Adversarial Attacks via Cross-Domain Interpretability}}, 
      author={Mahmoud Hossam and Trung Le and He Zhao and Dinh Phung},
      year={2020},
      Booktitle={25th International Conference on Pattern Recognition, (ICPR)}
}