PARSE: An Efficient Search Method for Black-box Adversarial Text Attacks

no code implementations • COLING 2022 • Pengwei Zhan, Chao Zheng, Jing Yang, Yuxiang Wang, Liming Wang, Yang Wu, Yunjian Zhang

Previous works on word-level attacks widely use word importance ranking (WIR) methods and complex search methods, including greedy search and heuristic algorithms, to find optimal substitutions.

Adversarial Text

Mitigating the Inconsistency Between Word Saliency and Model Confidence with Pathological Contrastive Training

no code implementations • Findings (ACL) 2022 • Pengwei Zhan, Yang Wu, Shaolei Zhou, Yunjian Zhang, Liming Wang

We show that the pathological inconsistency is caused by the representation collapse issue, which means that the representation of the sentences with tokens in different saliency reduced is somehow collapsed, and thus the important words cannot be distinguished from unimportant words in terms of model confidence changing.

Contrastive Learning

Multimodal Fusion with Co-Attention Networks for Fake News Detection

no code implementations • Findings (ACL) 2021 • Yang Wu, Pengwei Zhan, Yunjian Zhang, Liming Wang, Zhen Xu

Fake News Detection

360-Attack: Distortion-Aware Perturbations From Perspective-Views

no code implementations • CVPR 2022 • Yunjian Zhang, Yanwei Liu, Jinxia Liu, Jingbo Miao, Antonios Argyriou, Liming Wang, Zhen Xu

In this paper, we propose an adversarial attack targeting spherical images, called 360-attactk, that transfers adversarial perturbations from perspective-view (PV) images to a final adversarial spherical image.

Adversarial Attack