Environments. I am interested in creating open-ended, interactive environments that agents can explore and interact with. My work involves developing highly scalable and efficient representations of the world, particularly in 3D and 4D, together with methods for reconstructing and generating such environments.
Ang Cao
I am a Research Scientist at Google DeepMind. My research focuses on pushing the frontier of foundation model intelligence across digital worlds and, ultimately, the physical world.
Currently, I am particularly interested in discovering scalable, self-sustaining sources of learning signals that enable intelligent systems to improve beyond the limits of curated human supervision, including reinforcement learning, AI for AI (AI4AI), recursive self- and joint improvement, and learning from past experience.
Previously, I completed my Ph.D. at the University of Michigan, advised by Justin Johnson, and worked closely with Andrew Owens and JJ Park. I was also fortunate to intern at Meta FAIR and Meta GenAI.
News
Research interests
My research aims to advance foundation-model intelligence in digital worlds and, ultimately, the physical world. I approach this goal from three complementary perspectives: Environments, Intelligent systems, and Learning and continual improvement.
Intelligent systems. I study how to push the capability frontier of foundation models, particularly their ability to perceive, reason about, and act within their environments. I am especially interested in extending these capabilities from digital spaces to the physical world.
Learning and continual improvement. I am currently focused on how intelligent systems can continually improve beyond curated human supervision. I explore scalable, self-sustaining sources of learning signals, including interactions with environments, feedback from the system itself and from other models, and accumulated experience. My work spans reinforcement learning, AI for AI (AI4AI), and automated research (autoresearch), with an emphasis on recursive self-improvement and the joint improvement of agents and their environments.
ViLP
More work coming soon
|
