English(EN)Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges
大语言模型隐私研究聚焦日本数据、多模态风险与差分隐私适应性
作者PulseAugur 编辑部·[7 个来源]·
研究人员正在探索与大语言模型（LLMs）及其适应性相关的隐私风险。一项研究侧重于检测日本预训练语料库中的敏感个人信息，并开发了用于日本《个人信息保护法》下的特别关照个人信息（SCPI）的分类器。另一篇论文调查了多模态大语言模型中的隐私漏洞，强调了它们如何泄露图像和内存中的敏感数据，并引入了一个用于评估的数据集。第三项研究对差分隐私（DP）在适应大语言模型中的有效性进行了基准测试，发现数据分布的显著变化会影响隐私风险，而像LoRA这样的参数高效微调方法能为分布外数据提供更好的保护。
AI
arXiv:2606.12114v1 Announce Type: new Abstract: Sensitive personal information can appear in large-scale pre-training corpora for large language models (LLMs). Detecting and filtering such information is therefore essential to ensure compliance with privacy regulations and preven…
Sensitive personal information can appear in large-scale pre-training corpora for large language models (LLMs). Detecting and filtering such information is therefore essential to ensure compliance with privacy regulations and prevent unintended information leakage. However, in co…
Sensitive personal information can appear in large-scale pre-training corpora for large language models (LLMs). Detecting and filtering such information is therefore essential to ensure compliance with privacy regulations and prevent unintended information leakage. However, in co…
arXiv:2606.09125v1 Announce Type: cross Abstract: Privacy risks in text-only Large Language Models (LLMs) are well studied, particularly their tendency to memorize and leak sensitive information. However, Multi-modal Large Language Models (MLLMs), which process both text and imag…
arXiv cs.LG
TIER_1English(EN)·Bart{\l}omiej Marek, Lorenzo Rossi, Vincent Hanke, Xun Wang, Michael Backes, Franziska Boenisch, Adam Dziedzic·
arXiv:2606.09401v1 Announce Type: new Abstract: Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining…
Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining, where overlaps and interdependencies with adap…
Privacy risks in text-only Large Language Models (LLMs) are well studied, particularly their tendency to memorize and leak sensitive information. However, Multi-modal Large Language Models (MLLMs), which process both text and images, introduce unique privacy challenges that remai…