About

A major drawback of deep reinforcement learning (RL) is its poor data efficiency. In this tutorial, we present meta-RL as an approach to create sample-efficient and general-purpose RL algorithms, via learning the RL algorithm itself. Meta-RL aims to learn a policy that is capable of adapting to any new task from a distribution over tasks with only limited data. We present the meta-RL problem statement, along with an overview of methods, applications, and open problems on the path to making meta-RL part of the standard toolbox for a deep RL practitioner.

This tutorial is based on A Survey of Meta-Reinforcement Learning.

Information

• Tutorial date: September 13th, 2023

• Tutorial Time: 11:00 to 12:30 CEST (Berlin local time)

• Video: Tutorial Video

• Slides: Tutorial Slides 

Presenters

Authors

Syllabus

1. Introduction

2. Meta-RL Overview

3. Problem settings

4. Few-Shot Methods

   1. 1. • Parameterized policy gradients

         • Blackbox

         • Task Inference

5. Exploration

6. Results

7. Applications

8. Many-Shot Methods

   1. 1. • Meta-learned components

         • Outer-loop algorithms

9. Transfer to Novel Environments

10. Alternative supervision

11. Open problems