MARA
I worked on Autumn.cpp, the C++ interpreter for MARA’s modeling language, and on the MARA protocol for evaluating world-model learning. I’m now building world-modeling agents that combine LLMs with program synthesis.
People
Dat Nguyen
Research Scientist, Basis Research Institute
Dat Nguyen is a research scientist at Basis and a research associate in Nada Amin's lab at Harvard SEAS. He is broadly interested in the modeling of how we perceive the world, and the modeling of reasoning processes. To support this goal, he works in the emerging area between programming languages, machine learning, and probabilistic programming.
About
I am a research scientist at Basis and a research associate at Harvard’s Programming Languages and Formal Methods groups.
I am broadly interested in the modeling of how we perceive the world, and the modeling of reasoning processes. To support this goal, I work in the emerging area between programming languages, machine learning, and probabilistic programming.
I work on modeling uncertainty in symbolic world models in MARA, designing a robot design language that captures both morphology and control in R-ADA, and modeling LLM generation as an effect, as a framework for building agent harnesses, in Effectful.
I did my PhD at the University of Melbourne (2021–2024) on program repair and synthesis, and before that spent five years as an AI research engineer at Cinnamon AI Lab.
Projects
Current and recent Basis projects.
R-ADA
I’m helping build the formal language R-ADA uses to describe robots.
Collaborations
Ways to work with Dat at Basis.
Collaborative research projects
I’m happy to talk to people working on proof automation or neuro-symbolic methods, especially anyone trying to point those tools at real scientific problems.
Awards
Recognition for research contributions.
Best Paper, ARC Prize 2024
For “Combining Induction and Transduction for Abstract Reasoning.”
Software
Tools and research code.
Autumn.cpp
A C++ implementation of the Autumn modeling language, with bindings for WASM, Python, and Julia. Used inside MARA.
VRDSynth
Synthesizes programs that link entities across visually-rich documents. Replication code for the ISSTA 2024 paper.
NeuroSymbolicDG
Neuro-symbolic domain generalization via compositional layout grammars.
Recent Publications
Recent papers and preprints.
EMPIRIC: Experiment-Driven Learning of Residual World Models for Robot Planning
Y. Liang, A. Li, D. Nguyen, E. Bunnapradist, M. Naim, S. Kodali, Matteo Merler, Bowen Li, K. Gopinathan, Y. Liu, N. Pimpalkhare, Joshua B. Tenenbaum, Adrian Weller, Z. Tavares, Tom Silver, Kevin Ellis
The Scaffolding Paradox: Evidence from AI-Assisted Dafny Proof Synthesis
Anastasiya Kravchuk-Kirilyuk, D. Nguyen, Rhea Karty, Simon Henniger, Gabriel Poesia, Nada Amin
Benchmarking World-Model Learning with Environment-Level Queries
A. Warrier, D. Nguyen, M. Naim, Moksh Jain, Y. Liang, K. Schroeder, Cambridge Yang, Joshua B. Tenenbaum, Sebastian J. Vollmer, Kevin Ellis, Z. Tavares
VirDA: Reusing Backbone for Unsupervised Domain Adaptation with Visual Reprogramming
Duy Nguyen, D. Nguyen
ExoPredicator: Learning Abstract Models of Dynamic Worlds for Robot Planning
Y. Liang, D. Nguyen, Cambridge Yang, Tianyang Li, Joshua B. Tenenbaum, Carl Edward Rasmussen, Adrian Weller, Z. Tavares, Tom Silver, Kevin Ellis
Towards Reliable Evaluation of Neural Program Repair with Natural Robustness Testing
Thanh Le-Cong, D. Nguyen, Xuan Bach D. Le, Toby Murray
Combining Induction and Transduction for Abstract Reasoning
Wen-Ding Li, Keya Hu, Carter Larsen, Yuqing Wu, Simon Alford, Caleb Woo, Spencer M. Dunn, Hao Tang, M. Naim, D. Nguyen, Wei-Long Zheng, Z. Tavares, Yewen Pu, Kevin Ellis
Selected Publications
Earlier selected work.
VRDSynth: Synthesizing Programs for Multilingual Visually Rich Document Information Extraction
D. Nguyen, Tung Do-Viet, Hung Nguyen-Duy, Tuan-Hai Luu, Hung Le, Bach Le, Patanamon Thongtanunam
FFL: Fine-grained Fault Localization for Student Programs via Syntactic and Semantic Reasoning
D. Nguyen, Thanh Le-Cong, Duc M. Le, Xuan-Bach D. Le, Quyet-Thang Huynh
Articles
Basis essays and updates this person wrote or contributed to.
ExoPredicator: Abstracting Time and State for Robot Planning
We introduce ExoPredicator, a system that learns abstract world models for robot planning. By abstracting state, time, and both endogenous and exogenous causal processes, ExoPredicator enables robots to quickly learn how dynamic environments work and plan efficiently in them.
AutumnBench: World Model Learning in Humans and AI
We’re releasing a new version of Autumn with human baseline results, AI performance comparisons, and an interactive benchmark for world model discovery. This release includes the MARA protocol and provides a public platform for testing causal reasoning capabilities.