Christopher Ré
Last reviewed
Sources
17 citations
Review status
Source-backed
Revision
v1 · 1,677 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
17 citations
Review status
Source-backed
Revision
v1 · 1,677 words
Add missing citations, update stale details, or suggest a clearer explanation.
Christopher Re (his surname is sometimes written with an acute accent over the "e") is an American computer scientist and a professor of computer science at Stanford University, where he leads the research group known as Hazy Research. His work spans data management and machine learning systems, and several lines of it have become standard references: the DeepDive knowledge-base system, the Snorkel framework for weak supervision, the structured state space sequence model (S4) that seeded the Mamba architecture, and the FlashAttention algorithm. In 2015 he received a MacArthur Fellowship, the award commonly called a "genius grant." His research has also seeded a series of companies, several of which became billion-dollar businesses, among them Snorkel AI, SambaNova Systems, and Together AI. [1][2][3]
Re completed his undergraduate degree at Cornell University. [1] He then pursued graduate study in computer science at the University of Washington in Seattle, where he earned his PhD in 2009 under the supervision of the database researcher Dan Suciu. His doctoral dissertation, "Managing Large-scale Probabilistic Databases," addressed how to query and reason over data whose values are uncertain. The work was recognized with the ACM SIGMOD Jim Gray Doctoral Dissertation Award in 2010. [4][5]
After his PhD he joined the University of Wisconsin-Madison as an assistant professor, a position he held from 2009 to 2013. It was at Wisconsin that he began the DeepDive project on extracting structured knowledge from unstructured "dark" data. In 2013 he moved to Stanford, where he is a professor of computer science affiliated with the Stanford Artificial Intelligence Laboratory, the Machine Learning Group, and the Center for Research on Foundation Models. [1][2]
Re directs the Hazy Research group at Stanford. The group's stated interest is in the shifts brought about by machine learning and in building the foundations for next-generation machine learning systems, with a recurring focus on learning from increasingly weak forms of supervision and on designing algorithms that are aware of the hardware they run on. His research career divides roughly into an earlier body of work on data systems and a later body of work on machine learning architectures, though the two are connected by a consistent systems-oriented viewpoint. [2]
Re's early research centered on probabilistic databases and on systems for turning messy, unstructured inputs into clean, structured knowledge. The flagship system from this period was DeepDive, a framework for knowledge-base construction that used statistical inference to read documents, web pages, and other unstructured sources and populate a structured database of facts. DeepDive was applied to problems ranging from building scientific knowledge bases to assisting law-enforcement efforts against human trafficking, and it became the technical basis for one of Re's first companies. [2][6]
A defining theme of Re's machine learning research is weak supervision: training models without large hand-labeled datasets. With his students he introduced data programming, an approach in which users write labeling functions that encode heuristics, patterns, and rules, and a statistical model then estimates the accuracies of those noisy sources and combines them into probabilistic training labels. The idea was implemented in an open-source system called Snorkel, developed in his lab beginning around 2016, which let practitioners build training sets programmatically rather than by manual annotation. The data-programming paper appeared at NeurIPS in 2016 and the Snorkel system was described at VLDB in 2017. This line of work helped popularize what is now often called data-centric AI. [2][7][8]
In the foundation-model era, Hazy Research became known for rethinking the core building blocks of sequence models. Working with his PhD student Albert Gu, Re helped develop the structured state space sequence model, known as S4, described in the 2021 paper "Efficiently Modeling Long Sequences with Structured State Spaces" (with Albert Gu and Karan Goel). S4 reframed deep sequence modeling around classical state space models and set new results on long-range benchmarks, including tasks that earlier architectures could not solve. [2][9]
The S4 line of research led directly to Mamba, a selective state space model introduced in December 2023 by Albert Gu and Tri Dao, both of whom trained in Re's research orbit. Mamba offered a linear-time alternative to the attention mechanism of the transformer and became one of the most influential non-transformer architectures of the period. Re was not an author of the Mamba paper itself, but the model grew out of the state space model program he and his students had built. [10]
Re is also a co-author of FlashAttention, an input-output-aware algorithm that computes exact attention while dramatically reducing reads and writes to GPU memory. The 2022 paper, led by Tri Dao (whom Re co-advised with Stefano Ermon) along with Daniel Y. Fu, Atri Rudra, and Re, made it practical to train transformers on much longer sequences and was widely adopted across the industry. FlashAttention exemplifies the "hardware-aware" philosophy that runs through Re's later work: co-designing algorithms with the realities of modern accelerators. [11]
Beyond these headline results, Hazy Research has produced a range of related systems and models, including the Hyena long-convolution architecture, the ThunderKittens library for writing fast GPU kernels, and the Evo genomic foundation models for biology. The group's alumni have gone on to found or lead several AI companies, making the lab an unusually direct pipeline from academic research to commercial products. [2]
Re is a prolific entrepreneur whose companies have generally commercialized ideas first developed in his lab. By his own account he has co-founded a number of companies as well as a venture firm. [2] Two of his ventures were acquired by Apple: Lattice Data, which built on DeepDive to extract structure from unstructured data and was bought in 2017 in a deal reported at around 200 million dollars, and Inductiv, which built on the HoloClean automated data-cleaning system and was acquired in 2020. [12][13]
His other companies remained independent and several reached multi-billion-dollar scale. SambaNova Systems, founded in 2017 with Stanford colleague Kunle Olukotun and chief executive Rodrigo Liang, designs custom AI chips and full-stack systems as an alternative to mainstream GPUs and became one of the most heavily funded AI hardware startups. Snorkel AI, founded in 2019 with Alex Ratner (its chief executive) and other lab members, commercialized the weak-supervision approach and reached a one-billion-dollar valuation in 2021. Together AI, founded in 2022 with chief executive Vipul Ved Prakash, Ce Zhang, and Percy Liang, operates a cloud platform for training and serving open-source foundation models. In 2024 Re co-founded Cartesia with Albert Gu, Karan Goel, and others to commercialize real-time state space models. [14][15][16][17]
| Company | Founded | Focus | Notable outcome |
|---|---|---|---|
| Lattice Data (DeepDive) | 2015 | Extracting structure from unstructured "dark" data | Acquired by Apple, 2017 (reported ~$200M) |
| SambaNova Systems | 2017 | Custom AI chips and full-stack systems | Raised more than $1 billion in venture funding |
| Snorkel AI | 2019 | Data-centric AI and weak supervision | Reached a $1 billion valuation in 2021 |
| Inductiv (HoloClean) | circa 2018 | Automated data cleaning and repair | Acquired by Apple, 2020 |
| Together AI | 2022 | Cloud for open foundation models | Valued in the billions of dollars |
| Cartesia | 2024 | Real-time state space model AI | Venture-backed startup |
Re's best-known honor is the MacArthur Fellowship, awarded in 2015 and worth an unrestricted multi-year stipend; the foundation cited his work on tools that make it easier to extract knowledge from large, messy datasets. The University of Washington noted the award as a distinction for one of its computer science alumni. [3][5]
He has received a series of other major honors across both database research and machine learning. His Stanford profile lists, among others, a VLDB Early Career Award (2015), a Gordon and Betty Moore Data-Driven Discovery Award (2014), a SIGMOD Best Paper Award (2014), an Alfred P. Sloan Research Fellowship (2013), the Robert N. Noyce Faculty Fellowship (2013), a PODS Best Paper Award (2012), an NSF CAREER Award (2011), and the ACM SIGMOD Jim Gray Dissertation Award (2010). He also received an Okawa Research Grant in 2016. [2]
As of 2026 Re continues to lead Hazy Research at Stanford and remains active across his portfolio of companies, a body of work that has made his lab one of the most productive bridges between academic machine learning research and commercial AI systems. [2]