BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning to interact with humans | Timothy Lillicrap (DeepMind) - 
 Timothy Lillicrap (DeepMind)
DTSTART:20220428T160000Z
DTEND:20220428T170000Z
UID:TALK173252@talks.cam.ac.uk
CONTACT:74143
DESCRIPTION:Advances in deep reinforcement learning have allowed us to dev
 elop artificial agents that master intimidating problems such as Go\, Ches
 s\, and Starcraft. Yet\, we struggle to create agents that display general
  intelligence. We need new algorithms\, but it will be difficult to invent
  these in a vacuum. We need the right data streams to develop the next gen
 eration of algorithms. Progress in game playing was driven by tackling the
  end-to-end problem and training on the objective of interest. Since much 
 of what we call general intelligence is bound up in human problems and jud
 gements\, this suggests we should train agents through grounded interactio
 n with humans in open-ended settings. To lay groundwork for large-scale re
 inforcement learning with humans-in-the-loop\, we trained agents to imitat
 e the interactions of hundreds of human participants in a simulated enviro
 nment. Aided by self-supervised objectives\, our agents learn to respond t
 o humans in natural language\, following instructions and answering simple
  questions. While imperfect\, these agents offer a compelling starting poi
 nt for iterative interactive improvement via reinforcement learning on hum
 an feedback.
LOCATION:Lecture Theatre LT0\, Department of Engineering
END:VEVENT
END:VCALENDAR