Dan Roth (Hebrew: דן רוט) is the Eduardo D. Glandt Distinguished Professor of Computer and Information Science at the University of Pennsylvania and the Chief AI Scientist at Oracle. Until June 2024, he was a Vice President and distinguished scientist at AWS AI, where he led the scientific effort behind first-generation Generative AI products, including Titan Models, Amazon Q, and Bedrock, from inception to general availability. His research focuses on the computational foundations of intelligent behavior, particularly the integration of machine learning and inference methods for natural language understanding.
Roth received his B.A. summa cum laude in mathematics from the Technion in Israel and his Ph.D. in computer science from Harvard University in 1995. He taught at the University of Illinois at Urbana-Champaign from 1998 to 2017 before moving to the University of Pennsylvania.
Academic career and honors
Roth is a Fellow of the American Association for the Advancement of Science (AAAS), the Association for Computing Machinery (ACM), the Association for the Advancement of Artificial Intelligence (AAAI), and the Association of Computational Linguistics (ACL). His work has been recognized with multiple best paper awards and he has served on the editorial boards of major journals in artificial intelligence and NLP. He has also mentored numerous Ph.D. students who have gone on to academic and industry positions.
Research contributions
Roth's research centers on the idea that learning plays a central role in intelligence. He has made seminal contributions to the fusion of learning and reasoning, particularly through Constrained Conditional Models, which formulate NLP problems as integer linear programming (ILP) tasks. He also pioneered work on machine learning with weak, incidental supervision, allowing models to learn from noisy or indirect labels.
In 2008, Roth co-authored the first paper on zero-shot learning in NLP, introducing the concept of dataless classification, where models classify text without labeled examples by using external knowledge. He has also worked on probabilistic reasoning, including its computational complexity and probabilistic lifted inference, as well as part-based (constellation) methods in object recognition and response-based learning.
His developed tools for information extraction, including named entity recognition (NER), coreference resolution, wikification, semantic role labeling (SRL), and ESL text correction, are widely used in both academic research and commercial applications. These tools have influenced subsequent work in deep learning and large language models.
Industry leadership
At AWS AI, Roth oversaw the scientific direction for the company's generative AI initiatives, helping to bring Titan foundation models, the Amazon Q assistant, and the Bedrock platform to market. His leadership bridged academic research and practical deployment, ensuring that advances in transformer architectures and neural networks were translated into scalable cloud services.
In 2024, he joined Oracle as Chief AI Scientist, where he advises on the company's AI strategy and research agenda. He also co-founded NexLP, Inc., a startup applying NLP and machine learning to legal and compliance domains; NexLP was acquired by Reveal, Inc., an e-discovery software company, in 2020.
Current roles and advisory work
Roth serves on the scientific advisory board of the Allen Institute for AI, contributing to research on AI and NLP. He continues to publish actively and collaborate with researchers across institutions, focusing on topics such as curriculum learning, model pruning, and robust inference methods. His ongoing work aims to develop unified methodologies for intelligent behavior, integrating learning, reasoning, and language understanding.
Selected publications and impact
Roth has authored over 300 peer-reviewed papers, many in top venues such as ACL, NeurIPS, and AAAI. His 2008 dataless classification paper has been highly influential, anticipating later developments in zero-shot and few-shot learning. His research on learning and inference has also informed practical systems in information extraction and question answering, contributing to the broader field of natural language processing. He has given keynote talks at major conferences and has been a driving force in bridging statistical learning with symbolic reasoning.
References
- Source facts provided from Wikipedia (CC BY-SA).