adversarial

"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%。"

"Google DeepMind的研究人员系统化了自主智能体在浏览网页时可能遇到的新型攻击... [利用]信息环境本身作为武器。"

"人工智能安全现状：随着能力增长和模型可以监控其他模型，对抗性鲁棒性等问题依然存在，社会尚未为人工智能做好准备"

"我们的搜索策略通过在进化循环中避免对抗性训练来隔离架构对鲁棒性的影响。"

"我想开始一条关于这个的新研究线，使用微分几何学和动力系统等数学工具（还有其他建议吗？"

"我们的方法实现了更强的攻击效果，导致 AP50 下降超过 38%，同时比现有方法更好地保留了车辆结构并提高了人类感知的隐身性。"

"我们在皮质内语音解码上评估 ALIGN，发现它对以前未见过的会话的泛化能力始终更好，与基线相比，改进了音素错误率和单词错误率。"

"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。"

"对大型 RxRx1 和 RxRx1-WILDS 基准的广泛评估表明，ABRA 为 siRNA 扰动分类建立了新的技术水平。"

"大量实验表明，AOT 增强了 Defender 的感知鲁棒性并减少了幻觉，为训练更可靠的 MLLM 建立了可扩展的范例。"

"感觉教育渠道严重滞后。"

"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。"

"如果您厌倦了用缺乏真实世界 ctDNA 平均覆盖率和肿瘤突变负荷 (TMB) 变化的“噪声”的、经过消毒的公共领域数据来测试您的模型，我们应该谈谈。"

"Our findings demonstrate that static benchmarks inadequately predict safety under social pressure, necessitating multi-turn adversarial testing for clinical AI certification."

"Interpolating estimators must be suboptimal even under a subtle future $X$-attack, and achieving perfect fitting can substantially damage their robustness."

"Our study provides initial insights into the robustness and attack surface of diffusion language models."

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

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

""Claude is genuinely impressive, but the gap between 'looks right' and 'actually right' is bigger than I expected.""

"This paper introduces an Information-Obfuscation Reversible Adversarial Example (IO-RAE) framework, the pioneering method designed to safeguard audio privacy using reversible adversarial examples."

"Exploratory results demonstrated that ConvNeXt-Tiny achieved the highest performance, attaining a 96.88% accuracy on the test"

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

"The article likely discusses adversarial attacks and obfuscation techniques."

"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."

"Adversarial training is utilized to enhance user simulation for dialogue optimization."

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