Jiayuan Mao (Penn): “Integrated Learning and Planning with Neuro-Symbolic Concepts”
Abstract:
I aim to build intelligent agents that continually learn, reason, and plan: agents that can answer queries, infer human intentions, and pursue long-horizon goals spanning hours to days. In this talk, I will describe a general framework for learning and reasoning with neuro-symbolic concepts. Drawing inspirations from theories and studies in cognitive science, neuro-symbolic concepts serve as compositional abstractions of the physical world, representing object properties, relations, and affordances, and actions. Each concept combines symbolic programs, which specify how concepts can be structured and composed, with neural networks, which ground them in sensory inputs and agent actions. Like words combining into sentences, these concepts can be reused and recombined to support reasoning and planning in novel situations. I show that systems that leverage neuro-symbolic concepts demonstrate superior data efficiency, enable agents to reason and plan more quickly, and achieve strong generalization in novel situations and for novel goals.
Biography:
Jiayuan Mao is an assistant professor at the University of Pennsylvania, primary faculty in the General Robotics, Automation, Sensing & Perception (GRASP) lab, and lead of Lab One. She finished her PhD at MIT, advised by Josh Tenenbaum and Leslie Kaelbling.
Zoom: https://upenn.zoom.us/j/97645937545
