基本信息
浏览量:70
职业迁徙
个人简介
I have been doing research at the intersection of machine learning, systems, and policy, with a focus on auditing and improving machine learning systems’ compliance with policies, from the perspectives of
Privacy: I explore privacy risks and mitigation in distributed training [NeurIPS’21, NeurIPS’22, EMNLP-Findings’20, ICML’20] and retrieval-based language models [EMNLP’23]. I improve the efficiency [NeurIPS’23] and accuracy [ICLR’24] of differentially private training. My work has been deployed inside Google AI and Meta AI, resulted into an invited chapter in the textbook Federated Learning and a white paper on advancing Differential Privacy’s deployment in real-world applications.
Safety: I demonstrate safety alignment in existing large language models are brittle at the level of both behavior [ICLR’24] and knowledge [Preprint’24]. Addressing safety is crucial yet challenging. To promote dialogue and collaborative exploration of this critical issue, I am co-organizing the Princeton AI Alignment and Safety Seminar alongside Sadhika Malladi.
Data usage: I build tools to audit data usage in large language models [ICLR’24] and medical image analysis [IEEE TMI’22].
I also believe in the power of community efforts to enhance the trustworthiness and transparency of machine learning systems. Recently, we (with researchers from 13 institutes) advocate for A Safe Harbor for AI Evaluation and Red Teaming, encouraging AI companies to provide legal and technical protections for good faith research on their AI models. We also release an open letter (signed by 300+ researchers, and reported by The Washington Post, VentureBeat, AIPwn, and Computerworld).
Privacy: I explore privacy risks and mitigation in distributed training [NeurIPS’21, NeurIPS’22, EMNLP-Findings’20, ICML’20] and retrieval-based language models [EMNLP’23]. I improve the efficiency [NeurIPS’23] and accuracy [ICLR’24] of differentially private training. My work has been deployed inside Google AI and Meta AI, resulted into an invited chapter in the textbook Federated Learning and a white paper on advancing Differential Privacy’s deployment in real-world applications.
Safety: I demonstrate safety alignment in existing large language models are brittle at the level of both behavior [ICLR’24] and knowledge [Preprint’24]. Addressing safety is crucial yet challenging. To promote dialogue and collaborative exploration of this critical issue, I am co-organizing the Princeton AI Alignment and Safety Seminar alongside Sadhika Malladi.
Data usage: I build tools to audit data usage in large language models [ICLR’24] and medical image analysis [IEEE TMI’22].
I also believe in the power of community efforts to enhance the trustworthiness and transparency of machine learning systems. Recently, we (with researchers from 13 institutes) advocate for A Safe Harbor for AI Evaluation and Red Teaming, encouraging AI companies to provide legal and technical protections for good faith research on their AI models. We also release an open letter (signed by 300+ researchers, and reported by The Washington Post, VentureBeat, AIPwn, and Computerworld).
研究兴趣
论文共 27 篇作者统计合作学者相似作者
按年份排序按引用量排序主题筛选期刊级别筛选合作者筛选合作机构筛选
时间
引用量
主题
期刊级别
合作者
合作机构
Boyi Wei,Kaixuan Huang,Yangsibo Huang,Tinghao Xie,Xiangyu Qi,Mengzhou Xia,Prateek Mittal, Mengdi Wang,Peter Henderson
CoRR (2024)
引用0浏览0EI引用
0
0
CoRR (2023): 14887-14902
引用3浏览0EI引用
3
0
CoRR (2023)
引用0浏览0EI引用
0
0
ICLR 2024 (2023)
引用0浏览0EI引用
0
0
加载更多
作者统计
合作学者
合作机构
D-Core
- 合作者
- 学生
- 导师
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn