Course: MIT 6.S191: Introduction to Deep Learning

Lecture video: https://youtu.be/-WbN61qtTGQ

Lecturers: Alexander Amini and Ava Soleimany

Introduction

Deep reinforcement learning (RL) shifts from the paradigm where you have a learning model that is trained on a fixed dataset.

The algorithm is now going to be placed in a dynamic environment; it’s able to explore and interact with that environment in different ways; it can try different actions and experiences to learn how to best accomplish its task in that environment without human supervision or fixed annotations from humans.

You define an objective that the algorithm should try to optimize for.

This type of algorithm, known as reinforcement learning, has many applications in the real world, ranging from robotics to gameplay.

A popular example is an algorithm (called AlphaStar) built by DeepMind to play and compete with human players on StarCraft.