Master Reinforcement Learning: From Basics to Advanced Deep RL

What you will learn:

Define and understand Reinforcement Learning (RL)
Apply RL using OpenAI Gym, Stable Baselines, Keras-RL, and TensorFlow Agents
Identify key applications and success stories of RL
Compare and contrast Reinforcement Learning and Supervised Learning
Understand the main components of an RL problem setup
Identify the main ingredients of an RL agent and their taxonomy
Define and utilize Markov Reward Processes (MRP) and Markov Decision Processes (MDP)
Master different solution spaces within the MDP framework for RL
Solve RL problems using dynamic programming algorithms (policy evaluation, iteration, value iteration)
Solve RL problems using model-free algorithms (Monte Carlo, TD learning, Q-learning, SARSA)
Differentiate between on-policy and off-policy algorithms
Master Deep Reinforcement Learning (DRL) algorithms like Deep Q-Networks (DQN)
Master policy gradient algorithms and actor-critic methods (AC, A2C, A3C)
Master advanced DRL algorithms (DDPG, TRPO, PPO)
Define and understand model-based RL, differentiating it from planning

Description

Dive into the captivating world of Reinforcement Learning (RL) with our comprehensive online course! This course isn't just theory; it's a hands-on journey from foundational concepts to advanced Deep Reinforcement Learning (DRL) techniques. We'll demystify RL, comparing it to supervised learning and exploring its diverse applications in AI and machine learning.

Starting with the fundamentals, we'll build your understanding of the RL problem, its key components, and the Markov Decision Process (MDP). You'll master core algorithms like dynamic programming, Monte Carlo, Temporal Difference learning, Q-learning, and SARSA. Then, get ready to dive into the exciting realm of Deep Reinforcement Learning, where we'll cover Deep Q-Networks (DQN) and powerful policy-based methods such as Policy Gradients, A2C, A3C, TRPO, PPO, and DDPG. You'll gain practical experience using state-of-the-art libraries, including OpenAI Gym, Stable Baselines, Keras-RL, and TensorFlow Agents.

This course is structured into six key modules, each designed to build upon the previous one. You’ll complete practical exercises, work with real-world examples, and build your portfolio with projects on popular RL environments. By the end of the course, you'll be confident in applying RL algorithms and building intelligent agents for a wide range of applications.

Don't just learn RL—master it. Enroll today and unlock the power of reinforcement learning!

Curriculum

Introduction

This introductory module sets the stage for your RL journey. The "Course Introduction" and "Course Overview" lectures provide a comprehensive roadmap, laying the groundwork for the exciting topics to come.

Reinforcement Learning Fundamentals

This section lays the foundation. You'll define RL, understand its applications, and grasp the key differences between RL and supervised learning. Lectures cover the problem setup (AREA), reward functions, states, the components of an RL agent, policy, value, and models. You'll delve into OpenAI Gym environments for practical application and explore various RL agent taxonomies, differentiating between prediction and control problems.

Markov Decision Processes (MDP)

Here, you'll master the mathematical framework of RL. Lectures introduce Markov chains, Markov reward processes, and Markov decision processes. We’ll tackle prediction and control using Bellman equations and explore the action-value function Q, setting the scene for algorithmic solutions.

Solving MDPs: Algorithms and Techniques

This section explores various techniques for solving MDPs. You’ll learn about dynamic programming (policy evaluation, policy iteration, value iteration), Monte Carlo methods, Temporal Difference learning, SARSA, and Q-learning. Lectures focus on practical implementation and comparing on-policy and off-policy algorithms with illustrative examples.

Deep Reinforcement Learning (DRL)

This module introduces deep learning into the mix. We'll discuss how deep neural networks act as function approximators for value functions and policies. You’ll master Deep Q-Networks (DQN) and their application to large-scale problems, with practical exercises using Keras-RL and TF-Agents for Atari game solving.

Advanced DRL Algorithms and Applications

This advanced section covers policy-based methods, including policy gradients, REINFORCE, actor-critic methods (AC, A2C, A3C), TRPO, PPO, and DDPG. You'll learn to use the Stable Baselines library for implementation and explore real-world examples with Atari, Mario, and Street Fighter games.

Model-Based Reinforcement Learning

The final module explores model-based RL, contrasting it with planning techniques. We discuss model learning methods, sample-based planning, and the Dyna architecture, offering a comprehensive perspective on the RL landscape.

Conclusion

A concluding lecture summarises the key concepts and provides guidance for further learning. The "Material" section offers access to supplementary course slides.

Deal Source: real.discount