Papers by Shreyash Agarwal
“That Is a Suspicious Reaction!”: Interpreting Logits Variation to Detect NLP Adversarial Attacks (2022.acl-long)
Copied to clipboard
| Challenge: | Existing methods to detect adversarial text inputs are limited in performance and are not detectable via spell checkers. |
| Approach: | They propose a model-agnostic detector of adversarial text examples that detects patterns in the logits of the target classifier when perturbing the input text. |
| Outcome: | The proposed detector improves the state-of-the-art performance in recognizing adversarial inputs and exhibits strong generalization capabilities across different NLP models, datasets, and word-level attacks. |