arXivDaily arXiv每日学术速递 周一至周五更新

AI 大模型

大模型对齐与安全

大模型对齐、安全、越狱、红队、提示注入和可信评测。

2026-05-27 至 2026-05-27 共收录 3 信号源:cs.CL, cs.AI, cs.CY, cs.LG

1. 隐私与版权 3 篇

2603.13282 2026-05-27 cs.LG cs.AI 62%

FedTreeLoRA: Reconciling Statistical and Functional Heterogeneity in Federated LoRA Fine-Tuning

FedTreeLoRA:协调联邦LoRA微调中的统计异质性与功能异质性

Jieming Bian, Lei Wang, Letian Zhang, Jie Xu

机构 * University of Florida, Gainesville, FL 32611(佛罗里达大学) Middle Tennessee State University Murfreesboro, TN 37132(中田纳西州立大学)

专题命中 隐私与版权 :alignment(abstract);分类 cs.AI、cs.LG

AI总结 针对联邦LoRA微调中统计异质性与功能异质性正交耦合的问题,提出树结构聚合框架FedTreeLoRA,通过逐层对齐实现泛化与个性化的有效平衡。

Comments Accepted by ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.04711 2026-05-27 cs.CR cs.AI cs.LG 62%

SWAP: Towards Copyright Auditing of Soft Prompts via Sequential Watermarking

SWAP:通过顺序水印实现软提示的版权审计

Wenyuan Yang, Yichen Sun, Changzheng Chen, Zhixuan Chu, Jiaheng Zhang, Yiming Li, Dacheng Tao

机构 * Sun Yat-sen University(中山大学) Zhejiang University(浙江大学) National University of Singapore(新加坡国立大学) Nanyang Technological University(南洋理工大学)

专题命中 隐私与版权 :harmlessness(abstract);分类 cs.AI、cs.LG

AI总结 针对软提示的版权保护问题,提出一种基于顺序水印的审计方法SWAP,通过将水印嵌入到更复杂的输出分布顺序空间中,实现无害且鲁棒的版权验证。

Comments This paper has been accepted by the International Journal of Computer Vision (IJCV), 2026. The first two authors contributed equally to this work. 28 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.26882 2026-05-27 cs.CR 50%

Privacy-Preserving Screening for Record Linkage

隐私保护的记录链接筛选

Chenyu Huang, Fan Zhang, Huangxun Chen, Yongjun Zhao, Huaming Rao, Peng Chen, Danqing Huang

专题命中 隐私与版权 :alignment(abstract)

AI总结 提出 Screening-then-Linkage 框架,通过轻量级筛选阶段降低隐私保护记录链接的计算开销,并基于电路PSI设计Appraisal系统实现高效安全筛选。

Comments 14 pages; 2025 IEEE 41st International Conference on Data Engineering (ICDE)

Journal ref 2025 IEEE 41st International Conference on Data Engineering (ICDE)

详情

展开后加载摘要…

URL PDF HTML 收藏