Reinforcement Learning for Programmers (Author's Direct Lecture)

'Reinforcement Learning', a Core Technology for Future Business
We will explain the basic concepts in an easy and detailed manner. 🦾

■ Course Overview

This lecture was created based on the book “ Reinforcement Learning for Programmers . ” The author will personally teach the contents that could not be included in the paper . In 17 days, 2 hours a day , you can make reinforcement learning your own technology . From this moment on, reinforcement learning will not be a difficult and incomprehensible wall, but will be an excellent tool that you can freely use to increase your value .

The examples used in the lecture can be downloaded from the site https://github.com/multicore-it/rl .

■ Revised version of the lecture has been released.

The revised edition of 『Reinforcement Learning for Programmers』 has finally been released for those who hesitated to study reinforcement learning because of the mathematical theory and complex code. Through reinforcement learning, you will develop practical development skills that can create intelligent systems that can make judgments and adapt on their own in unpredictable situations. 🔗Shortcut

• Added more friendly and intuitive explanations.
• Added state-of-the-art practice tools (Stable Baselines3) and techniques (Optuna).
• We implemented a wealth of practical example projects (asset allocation strategy, branch rotation).

■ Why Reinforcement Learning?

Reinforcement learning is based on skill, not capital.

Reinforcement learning does not learn from pre-labeled data, but rather creates data by itself while running the agent, so there is less burden on data work and relatively less computing power is required . It is a field that can be disqualified because it depends a lot on a deep understanding of reinforcement learning algorithms and programming skills to solve problems.

Reinforcement learning is a key technology for future business innovation.

Reinforcement learning is an AI technology suitable for environments with limited capital, such as Korea . Many problems that arise in business environments can be solved with programming skills and reinforcement learning algorithms , and more advanced services and products can be created based on these characteristics .

■ Course Features

■ Learning Contents

In the section on basic concepts of reinforcement learning, we first explain the statistical and mathematical theories required for reinforcement learning, and then explain in detail the process from the MDP to the DQN algorithm .

In the section on artificial neural networks, rather than focusing on artificial neural networks, the process leading to artificial neural networks is explained step by step, starting from linear regression . Since it explains from the basics so that even people with no concept of artificial intelligence can understand , anyone with just a little bit of programming knowledge can easily understand .

In the value-based reinforcement learning section, the DQN algorithm is explained code-centrically . Among the various reinforcement learning algorithms, value-based reinforcement learning is relatively easy to understand, so it is introduced first .

In the policy-based reinforcement learning section , REINFORCE, A2C, and PPO algorithms are explained through code and guided through their direct execution . Policy-based algorithms are more difficult to understand than value-based algorithms, but they show relatively stable performance, so a lot of time is spent explaining them .

Finally, we explain reinforcement learning tuning . It covers everything from the detailed theory of artificial neural networks, which is essential for tuning , to Bayesian optimization techniques, which help efficiently tune algorithm parameters .

■ Program error measures

Please refer to the latest news "Program Error Action Guide (December 10, 2022)"