Ask Me Anything: Reinforcement Learning


Reinforcement learning is a way to train systems to learn how to make decisions and perform tasks using a reward function. To get started with RL, we’ll first walk through the basics of what it is and how it works, including some tips for getting started with reinforcement learning.

What is Reinforcement Learning?

Reinforcement learning is a type of machine learning that involves getting a computer to take actions in the world that maximize some reward. The reward can be anything, like winning points or collecting coins, but it’s usually something more abstract–like being close to accomplishing your goal.

The main difference between RL and other forms of machine learning is that instead of telling the computer what to do (e.g., “pick up object A”) we instead give it feedback about its progress toward achieving whatever goal we set out for it. This means that at any point during training, if you want your bot to try something else as opposed to repeating its current behavior over and over again until it finds success–say because doing so would take too long–you just tell it!

How does reinforcement learning work?

Reinforcement learning is a method of learning through trial and error. The system receives a reward for each action it takes, and can use the reward to learn which actions lead to the best rewards.

The goal of reinforcement learning is for your bot or agent to figure out how best to achieve its goals in any given situation. A simple example would be an AI program that’s learning how to play Sonic the Hedgehog (the classic Sega Genesis version). As you know from previous Ask Me Anythings, this game involves running through levels as fast as possible while collecting rings along the way–but there are bad guys who want nothing more than stopping our hero from achieving his destiny!

In order for our bot/agent/player character whatever we’ll call him here: “Sonic”–to succeed at reaching his destination at top speed while not getting killed by those evil robots along the way he needs some way of deciding what actions will allow him most effectively get past obstacles while also keeping himself safe from harm’s way…

What are the benefits of reinforcement learning?

Reinforcement learning is a powerful tool for building intelligent systems. It can be used to solve many problems, from teaching robots how to walk and talk, to making smart decisions in medical diagnosis and automated trading.

In short: RL is useful for learning from data (i.e., supervised learning), learning from experience (unsupervised), or even learning from other agents (reinforcement).

What are some of the challenges with RL?

  • Defining a reward function is often the hardest part of RL. It’s not enough to know that you want your agent to achieve a high score or win a game; how do you define what that means? Is it the number of points they have at any given time? How many points they have compared with other players? Are there other factors involved, like how much fun it was playing or if they enjoyed themselves? The answer depends on what kind of game you’re trying to model and how much information about its dynamics is available.
  • Defining loss functions can also be tricky. In some cases (such as games), there may be no clear way of measuring loss because it depends on factors outside our control such as other players’ behavior and luck rather than our own decisions; in others (like robotics), we might not have access to all available data needed for training–but more importantly: what would constitute failure!

How can I get started with reinforcement learning?

You can get started with reinforcement learning by reading papers, tutorials and books. There are also online courses available that will teach you the basics of reinforcement learning.

When you have a question about a specific aspect of reinforcement learning or its application to your use case, there are many communities where you can ask questions:

Reinforcement learning is a promising way to build smarter systems.

Reinforcement learning is a promising way to build smarter systems. It’s used for many different applications, including computer games and robotics.

Reinforcement learning has many applications: it can be used to train an agent (a computer system) how to behave in its environment or perform tasks that were previously not possible given our current technology. This type of machine learning has been successful in training agents how to play Atari games better than humans, drive cars autonomously and even make decisions on our behalf by making observations about the world around us!


We’ve covered a lot of ground here, and hopefully you have a clearer understanding of what reinforcement learning is and how it works. But this is just the beginning. As we continue to develop new algorithms that can solve complex problems with fewer resources, RL will become an increasingly important part of our everyday lives.