May 27, 2019 · To combat adversarial spelling mistakes, we propose placing a word recognition model in front of the downstream classifier.

Trained to recognize words corrupted by random adds, drops, swaps, and keyboard mistakes, our method achieves 32% relative (and 3.3% absolute) error reduction ...

To combat adversarial spelling mistakes, we propose placing a word recognition model in front of the downstream classifier. Our word.

This work proposes a word recognition model in front of the downstream classifier, outperforming both adversarial training and off-the-shelf spell checkers, ...

... The adversarial attacks considered include character-level manipulations such as swaps, substitutions, deletions, and insertions of significant words.

You can attack the already trained BiLSTM (word-only, char-only or word+char) models using swap/drop/key-board/add attacks. To do so use the following command.

May 27, 2019 · To combat adversarial spelling mistakes, we propose placing a word recognition model in front of the downstream classifier.

Combating Adversarial Misspellings with Robust Word Recognition. D Pruthi, B Dhingra, ZC Lipton. ACL 2019, 2019. 346, 2019 ; Learning to Deceive with Attention- ...

To combat adversarial spelling mistakes, we propose placing a word recognition model in front of the downstream classifier. Our word recognition models build ...

[docs]class Pruthi2019(AttackRecipe): """An implementation of the attack used in "Combating Adversarial Misspellings with Robust Word Recognition", Pruthi et al ...