Target-Agnostic Adversarial Attacks on Language Understanding Models
Published in arXiv preprint (2106.07047), 2021
Proposes adversarial attacks on language-understanding models that do not assume access to the victim’s label space or target outputs. The attacks transfer across tasks and demonstrate systemic robustness gaps in standard NLU model families.
Recommended citation: Chauhan, J., Bhukar, K., Kaul, M. (2021). "Target-Agnostic Adversarial Attacks on Language Understanding Models." arXiv preprint 2106.07047.
Download Paper