Fei Sha

Fei Sha is a machine learning researcher whose papers range from shallow parsing with conditional random fields to zero-shot learning and generative diffusion models for weather and climate. He earned a PhD at the University of Pennsylvania in 2007, worked at Yahoo! Research, and joined the University of Southern California in 2008, where he became a Sloan Research Fellow in 2013. After directing machine learning at Netflix Research in 2018, he joined Google Research in 2019, co-chaired AAAI-20 in 2020, and served as a program chair of ICLR 2025. He is an AI Research Scientist at Meta, with interests in probabilistic modeling and uncertainty quantification.

Now at: Meta
Badges: Bridge

Timeline

2003–2007

A PhD at Penn

He earns BSc and MSc degrees at Southeast University in Nanjing and publishes 'Shallow Parsing with Conditional Random Fields' with Fernando Pereira at the 2003 Human Language Technology Conference of the NAACL. At the University of Pennsylvania, he wins Outstanding Student Paper awards at ICML in 2004 and NIPS in 2006, then completes his PhD in 2007 under Lawrence K. Saul with a thesis on large margin training of acoustic models for speech recognition.
2007–2014

Assistant professor at USC

He is a research scientist at Yahoo! Research in Sunnyvale in 2007. In 2008, he joins the University of Southern California as an assistant professor, and in 2014 he becomes an associate professor.
2012–2013

A Sloan Research Fellowship

In 2012, he receives an Army Research Office Young Investigator Award. In 2013, the Alfred P. Sloan Foundation selects him as a Sloan Research Fellow, one of 126 researchers from the United States and Canada.
2016

Synthesized classifiers for zero-shot learning

In 2016, he coauthors 'Synthesized Classifiers for Zero-Shot Learning' with Soravit Changpinyo, Wei-Lun Chao and Boqing Gong. The method introduces 'phantom' object classes that serve as bases for building classifiers of unseen classes, and it is tested on four benchmark datasets, including ImageNet Fall 2011 with more than 20,000 unseen classes.
2018–2020

From Netflix to Google Research

He is Director of Machine Learning at Netflix Research in Los Gatos in 2018, then joins Google Research in Mountain View in 2019 as a research scientist and manager. At Google Research, he leads a team of scientists and engineers; in 2020, as a USC Viterbi professor, he is also one of two program co-chairs of the AAAI-20 conference, which draws about 8,000 full-length submissions.
2023–2024

Generative models for weather and climate

In 2023, he coauthors SEEDS with Lizao Li, Rob Carver, Ignacio Lopez-Gomez and John Anderson; the system emulates weather forecast ensembles with deep generative diffusion models learned from historical data. A 2024 paper, 'Dynamical-generative downscaling of climate model ensembles', written with Ignacio Lopez-Gomez, Zhong Yi Wan, Leonardo Zepeda-Núñez, Tapio Schneider and John Anderson, follows a regional climate model with a generative diffusion model to refine the resolution of climate projections.
2024–2025

Harvard associate and ICLR chair

In 2024, he becomes an associate at Harvard's School of Engineering and Applied Sciences. He is one of four Program Chairs of ICLR 2025, with Animesh Garg, Nanyun (Violet) Peng and Rose Yu.
As of October 2026

AI research scientist at Meta

He is an AI Research Scientist at Meta, after leading a team of scientists and engineers at Google Research. His interests include probabilistic modeling, uncertainty quantification, dynamical systems and probabilistic reasoning in LLMs.

Career

NowMetaAI Research Scientist
From 2024Harvard UniversityAssociate
From 2019Google ResearchResearch Scientist and Manager
2018–2019NetflixDirector of Machine Learning
2008–2014University of Southern CaliforniaAssistant Professor, Associate Professor
2007–2008YahooResearch Scientist