Kevin Ellis
Kevin Ellis is a computer scientist whose AI work explores how machines learn to write programs and reuse symbolic knowledge. He coauthored the program-synthesis systems DreamCoder and Write, Execute, Assess, completed a cognitive science PhD at MIT, and joined Cornell's computer science faculty. In 2024, he received an NSF CAREER Award to develop methods for symbolic learning, and his Cornell team won the ARC Prize's first-place paper award for a visual reasoning system. He and Saikat Dutta received a Meta grant in 2025 to build software development benchmarks. As of October 2026, he is an assistant professor of computer science at Cornell and continues research in program synthesis.
- Now at
- Cornell University
- Badges
Timeline
-
2015–2019
Program synthesis and a REPL
In 2015, he coauthors “Unsupervised Learning by Program Synthesis,” which combines probabilistic modeling with solver-based synthesis for visual learning and English morphology. In 2019, he coauthors “Write, Execute, Assess,” a neural program synthesis system that uses execution to search the space of possible programs.
-
2020
A cognitive science PhD
In 2020, he completes a PhD in cognitive science at MIT under Armando Solar-Lezama and Josh Tenenbaum. He also holds a B.S. in physics with a concentration in linguistics.
-
2020
DreamCoder learns program languages
In 2020, he coauthors DreamCoder, a system that learns to solve problems by writing programs. It creates programming languages for domain concepts and uses neural networks to guide program search.
-
2020–2021
From Common Sense Machines to Cornell
In September 2020, he is a research scientist at Common Sense Machines. In fall 2021, he moves to Cornell as an assistant professor of computer science, where his work includes program induction, model-structure learning, neurosymbolic methods and semantic parsing.
-
2024
An NSF CAREER Award
In 2024, he receives an NSF CAREER Award to develop AI methods that learn symbolic knowledge as computer code by combining statistics, large language models and program synthesis.
-
2024
A first-place ARC paper
In 2024, he coauthors “Combining Induction and Transduction for Abstract Reasoning” with his Cornell team; the paper wins the ARC Prize first-place paper award. Its two-model method combines program search with direct answer prediction and solves about 56% of the visual puzzles. Lead authors Wen-Ding Li and Keya Hu share the competition's $50,000 prize.
-
2025
Benchmarks for AI software
In 2025, he and Saikat Dutta receive a Meta grant to build benchmarks for generative AI in software development, including bug fixing, code review, test generation and programming-language translation.
-
As of October 2026
Assistant professor at Cornell
In June 2026, he is an assistant professor of computer science at Cornell and serves on the AI4AI research team. The initiative develops safety protocols for AI agents and their code. His research combines program synthesis with learning and reasoning from machine learning and cognitive science.
Career
- Since 2021Cornell UniversityAssistant Professor of Computer Science
- 2020–2021Common Sense MachinesResearch Scientist
Education
Sources and credits
Photo: We have no published photo of Kevin Ellis yet. Send us a free-licensed photo.
- Unsupervised Learning by Program Synthesis
- Write, Execute, Assess: Program Synthesis with a REPL
- Kevin Ellis | Department of Computer Science | Cornell Bowers
- Events
- DreamCoder: Growing generalizable, interpretable knowledge with wake-sleep Bayesian program learning
- Cornell Bowers CIS welcomes 13 faculty members
- Eight early-career professors win NSF development awards
- Cornell team wins $50K in AI puzzle-solving challenge contest
- Dutta and Ellis to advance AI coding with grant from Meta
- Amazon partnership establishes Cornell AI security initiative
- The Research Team | AI4AI


