Gradient-based Adversarial Attacks against Text Transformers

2021/04/15 by Chuan Guo, Guo, Chuan, Alexandre Sablayrolles +5 · 18 citations
Computer Science · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Cryptography and Security (cs.CR) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2104.13733

openalex publication_date 2021/04/15 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

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.

Citations

• Categorical Reparameterization with Gumbel-Softmax
• ALBERT: A Lite BERT for Self-supervised Learning of Language Representations

Cited by

• Safety Alignment of LMs via Non-cooperative Games
• MEEA: Mere Exposure Effect-Driven Confrontational Optimization for LLM Jailbreaking
• In-Context Representation Hijacking
• TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization
• MSCR: Exploring the Vulnerability of LLMs' Mathematical Reasoning Abilities Using Multi-Source Candidate Replacement
• SPARTA: Evaluating Reasoning Segmentation Robustness through Black-Box Adversarial Paraphrasing in Text Autoencoder Latent Space
• AutoPrompt: Automated Red-Teaming of Text-to-Image Models via LLM-Driven Adversarial Prompts
• SoK: Taxonomy and Evaluation of Prompt Security in Large Language Models
• ArenaBencher: Automatic Benchmark Evolution via Multi-Model Competitive Evaluation
• LatentBreak: Jailbreaking Large Language Models through Latent Space Feedback
• RL Is a Hammer and LLMs Are Nails: A Simple Reinforcement Learning Recipe for Strong Prompt Injection
• Influence-Guided Concolic Testing of Transformer Robustness
• Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings
• Adversarial machine learning :
• Semantic Representation Attack against Aligned Large Language Models
• Towards Inclusive Toxic Content Moderation: Addressing Vulnerabilities to Adversarial Attacks in Toxicity Classifiers Tackling LLM-generated Content
• Multi-task Adversarial Attacks against Black-box Model with Few-shot Queries
• Guided Perturbation Sensitivity (GPS): Detecting Adversarial Text via Embedding Stability and Word Importance

Related

• Step by Step Loss Goes Very Far: Multi-Step Quantization for Adversarial Text Attacks
• Block-Sparse Adversarial Attack to Fool Transformer-Based Text Classifiers
• A Differentiable Language Model Adversarial Attack on Text Classifiers
• Bridge the Gap Between CV and NLP! A Gradient-based Textual Adversarial Attack Framework
• GNP Attack: Transferable Adversarial Examples via Gradient Norm Penalty