Interpretable Adversarial Perturbation in Input Embedding Space for Text

Motoki Sato, Jun Suzuki, Hiroyuki Shindo, Yuji Matsumoto
5/8/2018
cs.LGcs.CLstat.ML

Official code: aonotas/interpretable-adv (42 stars, last commit 2020-02-27).

Abstract

Following great success in the image processing field, the idea of adversarial training has been applied to tasks in the natural language processing (NLP) field. One promising approach directly applies adversarial training developed in the image processing field to the input word embedding space instead of the discrete input space of texts. However, this approach abandons such interpretability as generating adversarial texts to significantly improve the performance of NLP tasks. This paper restores interpretability to such methods by restricting the directions of perturbations toward the existing words in the input embedding space. As a result, we can straightforwardly reconstruct each input with perturbations to an actual text by considering the perturbations to be the replacement of words in the sentence while maintaining or even improving the task performance.

View on arXivView PDF

Code Implementations(1)

aonotas/interpretable-advOfficial100%

Code for Interpretable Adversarial Perturbation in Input Embedding Space for Text, IJCAI 2018.

4213Shell, PythonMay 1, 20186 years ago
adversarial-textsinterpretablemachine-learningnlpvat

Similar Papers

T3: Tree-Autoencoder Constrained Adversarial Text Generation for Targeted Attack

84% match1 repo252019

Natural Language Induced Adversarial Images

84% match1 repo92024

FreeLB: Enhanced Adversarial Training for Natural Language Understanding

84% match1 repo2522019

Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailment

83% match1 repo5322019

VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models

83% match1 repo92024

Semantic-Aligned Adversarial Evolution Triangle for High-Transferability Vision-Language Attack

83% match1 repo422024

Cite this paper

@article{sato2018interpretable,
  title  = {Interpretable Adversarial Perturbation in Input Embedding Space for Text},
  author = {Motoki Sato and Jun Suzuki and Hiroyuki Shindo and Yuji Matsumoto},
  year   = {2018},
  eprint = {1805.02917},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1805.02917}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1805.02917.svg)](https://codeofpaper.com/paper/1805.02917)

Discussion