Asymptotically Optimal Agents

Tor Lattimore (ANU)

CS HDR MONITORING AI Group

DATE: 2011-04-27
TIME: 11:30:00 - 12:00:00
LOCATION: RSISE Seminar Room, A105 with Turkish Pide
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Artificial general intelligence aims to create agents capable of learning to solve arbitrary interesting problems. Unfortunately, what it means to be optimal as a general reinforcement learning agent has no clear definitions. I will present two versions of asymptotic optimality, one stronger than the other. I show that no agent can be strong asymptotically optimal and that there exist agents that are weak asymptotically optimal. I also demonstrate that such agents are necessarily incomputable.