NYU Tandon ECE Seminar Series on Modern AI : Discovering an agent's controllable latent state

Speaker: John Langford

Location: On-Line
Videoconference link: https://nyu.zoom.us/meeting/register/tJMpduCqqDgpGtAAnx1ih0XrAhj5j3sjif95?__s=xxxxxxx

Date: Wednesday, October 26, 2022

Given rich sensors (like cameras), how can you learn what an agent can do?  This is challenging in highly stateful environments where accomplishing tasks requires complex plans, since randomly exploring may never discover some capabilities of the agent.  It turns out however that some of the core principles of discovering controllable latent state are straightforward, enabling us to discover the controllable latent state after a few thousand steps even in environments where random exploration is not viable.  Furthermore, this forms a good foundational representation for optimizing reward functions.