World Models: A Multiverse We Can Act On
The agentic era needs more than imagined futures: models of actionable consequences, grounded neural–symbolic abstractions, and the worlds we share.
Read the storyNotes from the work
Blog
Research stories, lessons from building the academic community, and context that does not fit inside a paper abstract.
Download all blog citations
From Atoms to Bits: Building the Digital Development Loop for Physical Intelligence
Every physical trial should improve both the robot and the process that develops the next one.
Read the story
Self-Improving Agents: Learning How to Work
A self-improving agent should finish a task with more than a result: experience should change how it approaches the next one.
Read the story
Reasoning as Control: Toward Self-Improving Agentic Systems
Self-improvement belongs not only inside the model, but also in the infrastructure that controls thinking, action, evaluation, and workflow composition.
Read the story
Robotics Needs a Better Physical Language
Why the next generation of robot world models may depend less on bigger pixel predictors and more on compact, structured representations of interaction.
Read the story
When Students Shape the Science
Six graduating researchers, six intellectual arcs, and one evolving lab vision spanning data, governance, alignment, fairness, self-improvement, and physical AI.
Read the story
Where Has Furong Been? Behind the Scenes of Our NeurIPS Competition
The candid story of turning a research question about image-watermark robustness into a live NeurIPS competition—and the under-recognized service work behind it.
Read the story
NeurIPS ’22 Main Conference Papers from Furong Lab @ UMD
Six papers spanning fairness under distribution shift, robust reinforcement learning, distributed training, model invariance, and trustworthy machine learning.
Read the story