adversarial attacks

"我们展示了在DAN中平均攻击成功率（ASR）降低了28.2%，在WildJailbreak中降低了31.3%，在StrongREJECT基准测试中降低了35.4%。"

"RaPA (随机参数剪枝攻击) 能够显著提高对抗样本在不同模型之间的迁移攻击能力，也就是说，在一个模型上生成的攻击样本更容易欺骗其他模型。"

"该系统将基于梯度的对抗成功率从82.1%降低到18.7%，将时间抖动成功率从75.8%降低到25.1%，同时保持每次推理约45微焦耳的能耗。"

"为了实现更可靠的评估，我们提出了ReliableBench，这是一个行为基准，其判断结果更加一致，以及JudgeStressTest，这是一个旨在揭示判断失败的数据集。"

"在这项工作中，我们探讨了是否可以利用攻击信号来提高反事实解释的可靠性。"

"A small number of samples can poison LLMs of any size."

"By selectively flipping a fraction of samples from..."

"The research focuses on LLM-driven feature-level adversarial attacks."

"The paper focuses on time-efficient evaluation and enhancement."

"The article's context indicates it's a research paper from ArXiv, implying a focus on novel findings."

"The paper focuses on adversarial attacks against RF-based drone detectors."

"The article uses resume screening as a case study for analyzing adversarial vulnerabilities."

"The paper focuses on multi-layer confidence scoring for identifying out-of-distribution samples, adversarial attacks, and in-distribution misclassifications."

"The research is published on ArXiv."

"The research focuses on auditing soft prompt attacks against ESM-based variant predictors."

"An open-source testbed is provided for evaluating adversarial robustness."

"The research focuses on generalizing neural backdoor detection."

"The paper is available on ArXiv."

"The paper examines superposition, sparse autoencoders, and adversarial vulnerabilities."

"The research is published on ArXiv."

"The paper investigates defenses, economic impact, and governance evidence related to adversarial robustness in financial machine learning."

"The research focuses on adversarial attacks against deep learning-based radio frequency fingerprint identification."

"The paper focuses on evaluating Frank-Wolfe methods."

"The article's focus is on image protection methods for diffusion models."

"The research focuses on multi-turn adversarial attacks."

"The study reveals critical weaknesses of Vision-Language Models."

"The research focuses on automatic attack discovery."

"The context provided is very limited, so a key fact cannot be pulled."

"The article is likely about ways to 'fool' neural networks."

"The article is a short introduction, implying a high-level overview."