Overview

Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off

One annual plan covers every course and certificate on Coursera. 40% off for a limited time.

This course delves into advanced deep reinforcement learning (RL) algorithms, exploring state-of-the-art techniques such as DQN extensions, policy gradients, and actor-critic methods. It focuses on optimizing and extending RL models to address complex real-world tasks, making it essential for professionals working with AI in dynamic environments. Through a blend of theoretical discussions and practical applications, this course enables learners to apply RL strategies across domains like gaming, stock trading, and natural language environments. You’ll learn how to accelerate training processes and improve performance in diverse settings. By mastering these advanced RL algorithms, learners gain the ability to tackle complex challenges in various domains confidently. The course focuses on not just understanding the theory behind the algorithms but also implementing them effectively in practical scenarios. The course is perfect for professionals with a solid understanding of machine learning, especially those seeking to enhance their RL skills. Ideal for those working in AI development, game design, or financial modeling, it offers in-depth insights and actionable skills. This course is part two of a three-course Specialization designed to provide a comprehensive learning pathway in Reinforcement Learning. While it delivers standalone value, learners seeking an in-depth progression may benefit from completing the full Specialization.

Syllabus

DQN Extensions

This module explores advanced improvements to the Deep Q-Network (DQN) algorithm, including multi-step learning, noisy networks for enhanced exploration, prioritized replay buffers, and distributional approaches. Learners will gain practical experience implementing these extensions and analyzing their impact on training performance and efficiency.

Ways to Speed Up RL

This module explores practical strategies to accelerate reinforcement learning (RL) training, focusing on deep Q-network (DQN) improvements. Learners will investigate performance bottlenecks, experiment with batch sizes and parallelization, and understand the impact of environment wrappers on training efficiency. By the end, you'll be equipped to optimize RL workflows for faster convergence.

Stocks Trading Using RL

This module guides learners through applying deep Q-network (DQN) reinforcement learning techniques to real-world stock trading scenarios. You will work with historical Russian stock market data and explore different DQN architectures, including feed-forward and convolutional models, to develop and evaluate trading strategies.

Policy Gradients

This module introduces policy gradient methods as an alternative approach to solving Markov decision process problems in reinforcement learning. Learners will explore the mathematical foundations, implementation details, and practical considerations such as gradient variance and hyperparameter tuning. By working through real-world examples like CartPole, students will gain hands-on experience optimizing policies using neural networks.

Actor-Critic Methods - A2C and A3C

This module introduces policy-based reinforcement learning through actor-critic methods, focusing on A2C and A3C algorithms. Learners will explore how these methods reduce variance in policy gradients, implement parallel environments, and apply these techniques to classic control and Atari games. Practical coding exercises and performance analysis are included to solidify understanding.

The TextWorld Environment

This module introduces learners to solving text-based interactive fiction games using reinforcement learning within the TextWorld environment. You will explore game generation, deep NLP fundamentals, word embeddings, and preprocessing pipelines, culminating in training agents and integrating large language models like ChatGPT for automated gameplay. By the end, you'll understand how to process complex textual observations and apply RL techniques to dynamic, language-rich environments.

Web Navigation

This module explores how reinforcement learning can be applied to web navigation and browser automation tasks. Learners will experiment with simple RL agents in the MiniWoB environment, address challenges unique to browser automation, and enhance agent performance using text descriptions and human demonstrations.