Self- Racing Car and Reinforcement Learning

Original article was published by Christofel R Goenawan on Artificial Intelligence on Medium


What is Reinforcement Learning ?

Artificial Intelligence Fields. Image by Author.

Before we’re going to Reinforcement Learning, we should know what fields are there in Artificial Intelligence.

Reinforcement Learning is a learning type in Machine Learning fields, that is part of AI where Machine learns to solve a specific task by discovering pattern in the training phase.

There are 3 main sub- fields in Machine Learning fields :

  1. Supervised Learning

Supervised Learning is Machine Learning algorithm where we provided a lot of data and their corresponding outcome ( label ), and then the machine learns to discover relationship and pattern between data and the label ….

By learning the relationship and pattern between data and label the machine can predict the outcome of another given data.

Example of Supervised Learning is when we provided the financial information and loan history of some customers in certain Bank, and we provided whether the customer have bad credit or not. Then by using Supervised Learning, we can predict whether another customer will have bad credit or not based on their financial information and loan history.

Supervised Learning is the most used AI algorithm in the world right now.

2. Unsupervised Learning

Unsupervised Learning is Machine Learning algorithm where we provided a lot of data, but without their corresponding outcome. And then the machine learns to discover relationship and pattern between data.

By learning the relationship between data, the machine can discover pattern or relationship in data that are not so obvious for human. Hence help human to make a decision or gain actionable insight.

Example of Unsupervised Learning is when we provided the customer background information of a shop, like geography information, age, gender and their purchase history. Then by using Unsupervised Learning, we can clustering the customer background based on some background information and their corresponding purchase history, so we can understand our customer better and help us to design a marketing strategy or discount program.

3. Reinforcement Learning

Finally…. We’re Here….

What is Reinforcement Learning ( RL ) ?

Reinforcement Learning is Machine Learning algorithm where we try to make an entity ( called agent ) learns to take action in a specific environment in aim to maximize the cumulative rewards .

Example of Reinforcement Learning is when we train a chess player ( agent ) to play chess games ( environment ) in aim to win the game ( Rewards ), or when we train a car ( agent ) to drive in the street ( environment ) in aim to go to another place without crashing any people, other cars or objects ( Rewards ) .

Another great example of RL is when we train Mario to walk in Mario Bross Game. Let’s check this exciting application !!

Mario Bross using Reinforcement Learning by Gyfcat

In this example, we train Mario to walk in Mario Bross Game , where Mario try to avoid obstacles and get coins as much as he can. In this case, Mario as an agent learns to take action in Mario Bross Game environment , to maximize the number of coins he got ( Cumulative Rewards ).

Reinforcement Learning is Machine Learning algorithm where we try to make an agent learns to take action in a specific environment in aim to maximize the cumulative rewards

Greatt !! Now you know what is Reinforcement Learning….. But, How does it work ?