MARA
Modeling, Abstraction, and Reasoning Agents: systems that build and use world models through active experimentation and abstract reasoning.
People
Yichao Liang
Research Intern, Basis Research Institute
Yichao Liang is a research intern at Basis and a PhD student at the Computational and Biological Learning Lab (CBL) at Cambridge, supervised by Adrian Weller. His research focuses on how agents can learn modular, compositional abstractions of their environments and use them for reasoning, adaptation, and skill learning in open-ended settings, with the goal of building machine learning systems that are more sample-efficient, generalizable, and interpretable.
About
I am a research intern at Basis and a PhD student in the Computational and Biological Learning Lab at the University of Cambridge, supervised by Adrian Weller. My research lies at the intersection of machine learning, world models, planning and reasoning, and embodied AI. I study how agents can learn modular, compositional abstractions of their environments and use them for reasoning, adaptation, and skill learning in open-ended settings. My goal is to build machine learning systems that are more sample-efficient, generalizable, and interpretable. At Basis and Cornell, I work with Kevin Ellis, Zenna Tavares, and Tom Silver on neuro-symbolic approaches to world modeling and robot planning.Projects
Current and recent Basis projects.
R-ADA
A rational automated design agent for robotics, combining language models, simulation, probabilistic programming, and Bayesian inference.
Recent Publications
Recent papers and preprints.
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
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 Vollmer, Kevin Ellis, Z. Tavares
PoE-World: Compositional World Modeling with Products of Programmatic Experts
Wasu Top Piriyakulkij, Y. Liang, Hao Tang, Adrian Weller, Marta Kryven, Kevin Ellis
From Pixels to Predicates: Learning Symbolic World Models via Pretrained Vision-Language Models
Ashay Athalye, Nishanth Kumar, Tom Silver, Y. Liang, Jiuguang Wang, Tomás Lozano-Pérez, Leslie Pack Kaelbling
VisualPredicator: Learning Abstract World Models with Neuro-Symbolic Predicates for Robot Planning
Y. Liang, Nishanth Kumar, Hao Tang, Adrian Weller, Joshua B. Tenenbaum, Tom Silver, João F. Henriques, Kevin Ellis
Rapid Motor Adaptation for Robotic Manipulator Arms
Y. Liang, Kevin Ellis, João Henriques
Drawing out of Distribution with Neuro-Symbolic Generative Models
Y. Liang, Joshua B. Tenenbaum, Tuan Anh Le, N. Siddharth
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.