Talk / Overview

Reinforcement learning (RL) typically assumes a well-specified reward function to be available, but this is rarely the case in practical settings. Instead, this talk will discuss how RL agents can learn what to do from human feedback if no standard reward signal is available. In particular, we focus on learning robust representations of tasks in a sample-efficient way.

Talk / Speakers

David Lindner

Doctoral Student, ETH Zurich

Talk / Slides

Download the slides for this talk.Download ( PDF, 4388.48 MB)

Talk / Highlights

Learning Task Specifications for Reinforcement Learning from Human Feedback

With David LindnerPublished May 02, 2022

AMLD / Global partners