[2104.13733] Gradient-based Adversarial Attacks against Text Transformers

We propose the first general-purpose gradient-based attack against transformer models. Instead of searching for a single adversarial example, we search for a distribution of adversarial examples parameterized by a continuous-valued matrix, hence enabling gradient-based optimization. We empirically demonstrate that our white-box attack attains state-of-the-art attack performance on a variety of natural language tasks. Furthermore, we show that a powerful black-box transfer attack, enabled by sampling from the adversarial distribution, matches or exceeds existing methods, while only requiring hard-label outputs.

Gradient-based Adversarial Attacks against Text Transformers Chuan Guo∗ Alexandre Sablayrolles∗ Hervé Jégou Douwe Kiela Facebook AI Research Abstract and L∞ -norms, but such metrics are not readily ap- plicable to text data. To circumvent this…

related reading

• A Mathematical Framework for Transformer Circuitstransformer-circuits.pub
• transformer_attention.pdfarxiv.org
• [2307.15043] Universal and Transferable Adversarial Attacks on Aligned Language Modelsarxiv.org
• Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentarxiv.org
• [2212.07677] Transformers learn in-context by gradient descentarxiv.org
• 2306.15447.pdfarxiv.org
• Adversarial Attacks on LLMs | Lil'Loglilianweng.github.io
• [2005.11401] Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasksarxiv.org
• [2602.04899] Phantom Transfer: Data-level Defences are Insufficient Against Data Poisoningarxiv.org
• [2106.09685] LoRA: Low-Rank Adaptation of Large Language Modelsarxiv.org
• [2403.06634] Stealing Part of a Production Language Modelarxiv.org
• 1706.03762arxiv.org