Prof. Jindong Wang is recruiting fully funded PhD students

[Chinese version]

Dr. Jindong Wang is currently an assistant professor at William & Mary, one of America’s “Public Ivies,” since January 2025, and also serves as an affiliated faculty member at the renowned non-profit Future of Life Institute. From 2019 to 2024, he was a Senior Researcher at Microsoft Research Asia. Prospective PhD students interested in joining his group are welcome to apply. Research areas: foundation models and machine learning, the philosophy of language models, and the intersection of AI and the social sciences.

He has published 60+ papers in top venues such as ICML and NeurIPS, with 29,000+ citations (h-index 62). He has been named among Stanford University’s global top 2% scientists for four consecutive years, and his work has been covered by MIT Technology Review and Forbes. He maintains close ties with leading universities and companies. Within less than a year of joining W&M, he has received research awards and funding from Google, Amazon, Microsoft, AMD, Cohere, and others. The lab has ample GPU and API resources. He has supervised 10+ undergraduate, master’s, and PhD students to publish top-tier papers starting from scratch. He is an Associate Editor of IEEE TNNLS and JCST, and serves as Area Chair for ICML, NeurIPS, ICLR, KDD, and ACL.

Personal site: jd92.wang. Application form: forms.gle/uKTAz3n9ySPeemMBA.

University and City

Founded in 1693, William & Mary is the second-oldest university in the United States after Harvard and is known as the “Alma Mater of the Nation.” As a “Public Ivy,” it consistently ranks among the top institutions nationwide and is an R1 research university. It has produced many distinguished alumni, including three U.S. Presidents: Thomas Jefferson, James Monroe, and John Tyler. The university is known for its small size and excellence, favorable student–faculty ratio, and rich academic atmosphere. In the 2026 U.S. News rankings it is 21st among public universities and 51st overall, and it is one of the public universities most recognized by alumni and parents.

• Location and climate: The university is located in Williamsburg, Virginia, a historic city and vacation destination on the U.S. East Coast. It is safe and comfortable, with four distinct seasons and pleasant weather. The beach is 30 minutes away. Winters and summers are moderate rather than extreme. The natural scenery is beautiful and the community is harmonious, making it ideal for study and life. A rough analogy in China would be Qingdao or parts of northern Jiangsu.
• Living and shopping: Although the city is small, it has many conveniences: major supermarkets (Walmart, Target, Publix, Harris Teeter, Trader Joe’s), Premium Outlets, cinemas, cafés, Chinese restaurants, and diverse international cuisine. Busch Gardens and Water Country are the largest theme parks on the U.S. East Coast, offering plenty of leisure options.
• Transportation and travel: Nearby larger cities include Newport News (30 minutes), Norfolk (50 minutes, home to the largest U.S. naval base), Richmond (50 minutes, the capital of Virginia), Virginia Beach (1 hour), and Washington, D.C. (2.5 hours). It is a 50-minute drive to Richmond International Airport and 2.5 hours to Washington airports. Students can enjoy the tranquility of a small town with convenient access to major metropolitan areas.
• History and culture: Williamsburg is one of the most famous historic cities in the U.S., home to Colonial Williamsburg, the world’s largest living-history museum. Students can experience 18th-century American life and study the history of independence and democracy. Nearby are many museums such as the Williamsburg Art Museums, Virginia Museum of Fine Arts, The Mariners’ Museum in Newport News, and the American Revolution Museum.

Research Areas and Mentoring Experience

Mentoring: During his time at Microsoft, Prof. Wang supervised 10+ interns who published their first top-tier conference papers, averaging roughly one top-tier paper per person per six months. Several students had already surpassed 3,000 Google Scholar citations before graduating. He respects students, avoids micromanagement, and is committed to helping them grow. Under his guidance, some students pursued PhDs at UW and UCSB, or joined top companies like Google DeepMind and Microsoft. PhD projects in the group are conducted in communication and collaboration with industry partners and senior PhD students. He is active on social media, sharing insights on research and life over many years.

Representative work: Before the LLM era, he published widely in core machine learning areas such as transfer learning, OOD generalization, semi-supervised learning, and federated learning. Representative works include transfer learning algorithms MEDA (800 citations) and DSAN (1,200 citations in 5 years), semi-supervised algorithms FlexMatch (1,300 citations in 4 years) and FreeMatch (500 citations in 2 years), and the federated learning algorithm FedHealth (1,100 citations in 5 years). In the era of large models, he has actively embraced new technologies and made solid progress in evaluation, alignment, fine-tuning, and agents. Representative works include the dynamic evaluation protocol DyVal and the evaluation framework PromptBench, the agent framework CompeteAI, the new direction of noisy model learning, and the psychology-inspired EmotionPrompt. Below are areas and outputs from the past two years.

• Machine learning and foundation models: Making foundation models more robust, trustworthy, and responsible. Topics include robust ML, OOD generalization, transfer learning, semi-supervised learning, and federated learning. Recently, he has been especially interested in ML based on large foundation models, such as studying the catastrophic inheritance problem of pretraining data (vision paper (DMLR’24)).
  • Related papers: Noisy model learning (ICLR’24 spotlight), Noisy diffusion pre-training (NeurIPS’24 spotlight), Diversify (TPAMI’24).
• Philosophy of language models: Understanding the potential and limitations of large foundation models. Exploring why they work, their capability boundaries, risks, alignment, agents, and more.
  • Related papers: DyVal (ICLR’24 spotlight), DyVal2 (ICML’24), PromptBench (JMLR’24), PromptRobust (CCS LAMPS), Personalized Safety (NeurIPS’25).
• Generative AI and social sciences: How do we measure the impact of generative AI across domains? How can powerful AI models aid interdisciplinary research? How can existing social science knowledge help us better understand AI models?
  • Related papers: CompeteAI (ICML’24 oral), EmotionPrompt (ICML’24), CultureLLM (NeurIPS’24), CulturePark (NeurIPS’24), AgentReview (EMNLP’24 Oral).

Computing Resources

The group currently has 2 PhD students and 3 remote interns, with the following resources:

• Dedicated lab GPU cluster (4×H100, 8×L40S, etc.)
• Sponsored GPUs from industry partners (Google, Amazon, AMD, Modal, etc.)
• Sufficient GPT and Gemini API credits

Academic and Industry Collaboration

We collaborate widely with institutions across the U.S., Europe, Australia, Japan, Korea, Hong Kong, and Singapore, including but not limited to: Carnegie Mellon University, Stanford University, Duke University, UIUC, University of Wisconsin–Madison, UCLA, UCSB, Georgia Tech, UIC, Max Planck Society (Germany), Idiap (Switzerland), Nanyang Technological University, University of Technology Sydney, The University of Hong Kong, The University of Tokyo, KAIST, Westlake University, Peking University, and others. Prof. Wang maintains strong collaborations with both established and rising scholars and fully supports co-advising with these mentors.

We also work closely with leading companies such as Microsoft, Google, Amazon, and NVIDIA, providing students with advantages for subsequent collaborations and internships.

PhD Applicant Requirements

• First and foremost: good character, earnest attitude, and genuine sincerity toward advisor, peers, and collaborators
• Strong self-motivation to conduct high-impact research
• Excellent understanding of machine learning and AI, with strong mathematical foundations
• Strong programming, writing, and English skills (TOEFL 100 or equivalent)
• Solid transcripts and academic record
• Three recommendation letters

How to Apply

• For outreach (responses not guaranteed):
  • Fill out the online form: forms.gle/uKTAz3n9ySPeemMBA
  • I don’t accept email applications now due to high volumes of emails. I only accept the Google Form. Don’t send me emails.
• Apply through the university website:
  • Visit the William & Mary Ph.D. in Data Science page and submit your application.

FAQ

Q: What is the graduation requirement for PhD students in your group?
A: At least three high-quality papers that can support a PhD dissertation. No hard time limits. Early graduation is encouraged.

Q: Do you support internships in industry or at other universities?
A: Absolutely, as long as you are working on the right things.

Q: Do you accept (remote) interns or research assistants and write recommendation letters?
A: Yes, as long as you do your best.

Q: Do you select students based on background factors such as nationality, university, or race?
A: I did not come from a top university myself. I value your attitude and ability, not other factors.

Q: Do you put pressure on students?
A: No. You can ask my former interns about my reputation. I prefer that you push yourself, and we will work hard together.

Q: Do you support students attending academic conferences?
A: Yes. My start-up funds include budget for this, and there will be more support in the future.

Q: Do you accept co-advising with other professors?
A: Yes. We are actively exploring more collaborations with top universities.

Q: What kinds of applicants are discouraged from applying?
A: Anyone who meets any of the following: (1) Places excessive emphasis on external factors like rankings, advisor pedigree and connections, or domestic recognition by friends and relatives. (2) Strongly prefers big-city life and cannot tolerate the tranquility of a small town. (3) “Collects offers and disappears.” (4) Wants to coast through a PhD without putting in real effort.

中文版本

王晋东博士于2025年1月加入美国“公立常春藤”之一的威廉玛丽学院担任助理教授，同时在著名非盈利机构 Future of Life Institute 兼任导师。他于2019–2024年在微软亚洲研究院担任高级研究员。本招生广告长期有效。研究方向：基础模型与机器学习、语言模型的哲学、以及多模态模型与智能体。

他在 Nature、Nature 子刊、ICML、NeurIPS、ICLR 等顶级会议和期刊发表了60余篇论文，被引超过29,000次（H指数62），连续4年被斯坦福大学评为全球前2%科学家（本校 CSRanking 排名第一），成果被《麻省理工科技评论》和《福布斯》等知名媒体报道。他与顶尖高校和企业密切联系，入职不到一年已获得谷歌、亚马逊、英伟达、微软、AMD、Cohere 等企业研究奖项与资金支持，拥有充足的 GPU 和 API 资源。他已指导10余位本科、硕士和博士生从零起步发表顶级文章。他现任 IEEE TNNLS 和 JCST 的副主编，并担任 ICML、NeurIPS、ICLR、KDD 和 ACL 等的领域主席。

个人主页： jd92.wang 。申请链接： forms.gle/uKTAz3n9ySPeemMBA 。

大学和城市介绍

威廉玛丽学院（William & Mary）始建于1693年，是仅次于哈佛的美国第二古老大学，被誉为“美国母校”。作为“公立常春藤”之一，它长期位于全美顶尖学府之列（R1研究型大学），培养了包括托马斯·杰斐逊、詹姆斯·门罗和约翰·泰勒三位美国总统在内的大批杰出校友。学校以小而精著称，师生比优越，学术氛围浓厚。2026年 US News 排名全美公立大学第21位、综合第51位，是美国最受校友和家长认可的公立大学之一。1995年江泽民主席访美时在美国本土的第一站便是威廉斯堡。（我整理的威廉斯堡吃喝玩乐汇总）

• 位置与气候： 学校坐落于美国东海岸弗吉尼亚州的威廉斯堡，是一个历史名城和度假胜地。这里安全舒适、四季分明、气候宜人，离海边半小时、冬夏没有极冷极热、自然风光优美、居民和谐，非常适合学习和生活（类比国内青岛或苏北地区）。
• 生活与购物： 城市虽小，却拥有众多便利设施：大型超市（Walmart、Target、Publix、Harris Teeter、Trader Joe’s）、奥特莱斯购物中心（Premium Outlets）、电影院、咖啡馆、中餐馆和国际餐厅一应俱全。著名的 Busch Gardens 和 Water Country 是美国东部最大的主题乐园，为休闲娱乐提供丰富选择。
• 交通与出行： 学校临近周边更大城市，如 Newport News（半小时）、Norfolk（50分钟，美国最大的海军基地所在地）、Richmond（50分钟，弗吉尼亚州州府）、Virginia Beach（1小时，弗吉尼亚海滩度假胜地）、DC（2.5小时，美国首都华盛顿）。距离里士满国际机场50分钟车程、距离华盛顿机场2.5小时车程。学生既能享受小城的安宁，又能便捷通达大都市。
• 历史与文化： 威廉斯堡是美国最著名的历史文化名城之一，拥有全球最大的活历史博物馆——殖民地威廉斯堡。学生不仅能感受18世纪的美国风貌，还能深度学习美国独立和民主的历史。周边有众多博物馆，如威廉斯堡艺术博物馆、弗吉尼亚美术馆、新港纽斯海军博物馆、美洲革命博物馆等。

研究方向与以往指导经验

指导学生经验： 王晋东老师在微软期间指导10余名实习生发表人生中的第一篇顶级会议论文，平均每人每半年发表一篇顶会文章，多人毕业前谷歌被引已突破3000次。他尊重学生、不微操，致力于帮助他们成长。指导学生成果：顶级高校读博（华盛顿大学、北京大学、加州大学圣巴巴拉分校、港中文等），顶级企业就业（Google DeepMind、微软、AMD等），顶级企业实习（Anthropic Fellow、月之暗面、苹果、腾讯等），在 William & Mary 指导的本科生获得暑期研究奖、优秀论文奖等。组内 PhD 学生的研究项目均与大企业和高年级 PhD 沟通合作。他活跃于知乎、X、小红书和 B站等平台，多年来持续分享对于研究和生活的见解。

代表工作： 在大模型之前，他在机器学习、迁移学习、OOD泛化、半监督学习、联邦学习等基础机器学习方向发表了大量成果，代表作如迁移学习算法 MEDA（被引800次）和 DSAN（5年被引1400次）、半监督学习算法 FlexMatch（4年被引1600次）和 FreeMatch（2年被引800次），及联邦学习算法 FedHealth（5年被引1400次；联邦健康领域最高引技术文章）；在大模型时代，他积极拥抱新技术，在大模型的评测、对齐、微调、智能体等方面也取得良好进展，代表工作包括动态评测协议 DyVal 与评测开源框架 PromptBench、智能体 CompeteAI、新方向 noisy model learning，以及心理学大模型方法 EmotionPrompt（福布斯报道）。以下为近2年的研究方向与成果。

• 大语言模型：评测、对齐与智能体： 研究大语言模型的能力、行为与局限，重点关注如何科学地评测模型的知识与推理能力，如何实现个性化、安全与价值对齐，以及如何构建具有推理、规划和协作能力的智能体系统。
  • 相关主题： AI Evaluation Science、AI Personalization & Alignment、AI Safety、Agentic AI、Multi-Agent Systems
  • 相关论文： ADELE (Nature)、Classroom AI (Nature NPJ AI)、KnowledgeSmith (ICLR’26)、SparseEval (ICLR’26)、Personalized VLM safety (COLM’26)、Personalized Safety (NeurIPS’25)、CultureLLM (NeurIPS’24)、CulturePark (NeurIPS’24)、AgentArk。
• 统一多模态基础模型：理解、生成与推理： 研究如何在一个统一模型中融合文本、图像等多种模态的理解、生成与推理能力，重点探索理解与生成之间的一致性、统一表示与学习机制、多模态推理，以及模型的高效训练、适配与协同。
  • 相关主题： Unified Multimodal Models、Multimodal Reasoning、Understanding-Generation Consistency、Latent Alignment、Multimodal Adaptation
  • 相关论文： UniGame (CVPR’26)、xLARD (CVPR’26)、UMM (AAAI’26)、FairUMM (NeurIPS’25)、TorchUMM、UniPath、LatentUMM、FedUMM (WWW’26)。
• 可信基础模型：鲁棒性、泛化与错误继承： 研究预训练数据和基础模型中的噪声、偏差与错误如何被下游模型继承甚至放大，并探索利用基础模型提升机器学习系统的鲁棒性、OOD 泛化能力、可靠性和安全性。
  • 相关主题： Catastrophic Inheritance、Robust Learning、OOD Generalization、Noisy Pre-training、Bias Inheritance、Trustworthy ML
  • 相关论文： Catastrophic Inheritance (DMLR’24)、Noisy Model Learning (ICLR’24 Spotlight)、Noisy Diffusion Pre-training (NeurIPS’24 Spotlight)、Bias Inheritance (ACL’26 Oral)、HAROOD (KDD’26 Oral)、Masked Autoencoder (ICML’25 Spotlight)。

计算资源

目前组里有2个 PhD 学生和3个（远程）实习生，配备以下资源：

• 组里专用的 GPU 集群（4 H100，8 L40S，4 RTX Pro 6000等）
• 企业合作伙伴赞助的 GPU（Google、Amazon、AMD、Modal等）
• 充足的 GPT 和 Gemini API
• NSF共享GPU：NCSA DeltaAI（608 NVIDIA H100 GPUs）

学术和产业合作

我们与美国、欧洲、澳洲、日本、韩国、香港、新加坡等全球机构保持广泛合作，包括但不限于：卡内基梅隆大学、斯坦福大学、杜克大学、伊利诺伊大学厄巴纳-香槟分校、威斯康辛麦迪逊、加州大学洛杉矶分校、圣巴巴拉分校、佐治亚理工、伊利诺伊大学芝加哥分校、德国马普所、瑞士 Idiap 研究所、芬兰奥卢大学、新加坡南洋理工大学、悉尼科技大学、香港大学、东京大学、韩国科学技术院、西湖大学、北京大学等。他与知名教授和年轻学者均保持良好合作关系，并全力支持与这些导师的联合指导。

我们也与微软、谷歌、亚马逊、英伟达等企业深入联系，为学生提供了合作和实习的优势。

博士生要求

• 首要标准： 品德良好，态度认真，对导师、同学、合作伙伴真诚不虚伪
• 具有强烈的自我驱动力开展高影响力研究、研究方向与我匹配
• 对机器学习和人工智能有出色的理解；出色的数学功底
• 优秀的编程、写作和英语技能（托福100分或对等的语言成绩）
• 良好的成绩单和学习记录
• 三封推荐信

如何申请

• 套磁专用（不保证回复）：
  • 在线填写表格：forms.gle/uKTAz3n9ySPeemMBA
  • 请不要直接给我发邮件或试图通过其他社交媒体联系，没时间回复。仅接受 Google Form 套磁。
• 在大学官方网站申请：
  • 访问 William & Mary 数据科学博士项目页面 并提交申请。

常见问题

问：您研究组的博士毕业标准是什么？
答：至少发表三篇高质量论文，能够支撑一篇博士论文。不卡年限、鼓励提前毕业。

问：您支持在业界或其他大学实习吗？
答：当然支持！只要你在做正确的事情。

问：您接受（远程）实习生或研究助理并写推荐信吗？
答：接受。只要你尽最大努力做好事情。

问：您是否根据学生的背景（如国籍、大学或种族）来挑选学生？
答：我并非来自顶尖大学。因此，我只看重你的态度和能力，而不是其他任何因素。

问：您会给学生施加压力吗？
答：不会。你可以向我之前的实习生了解我的口碑，我自己就不王婆卖瓜了。我希望你自己鞭策自己，我们一起努力。

问：您支持学生参加学术会议吗？
答：是的。我的启动资金中有这部分预算，未来还会获得更多支持。

问：您接受与其他教授共同指导吗？
答：可以。我们也在积极探索与更多名校教授合作。

问：不建议什么样的学生申请？
答：满足以下任意一条即不建议申请：(1) 对学校排名、导师出身背景和 connection、国内亲戚朋友认可度等外在因素极为看重；(2) 受不了小城市宁静生活、极度向往大城市；(3) 拿 offer 就跑的“海王”；(4) 想躺平混个博士学位的学生。