Reinforcement Learning Part 0

This Post first appeared Here

We at init 27 labs are super excited to announce an upcoming Reinforcement Learning Series!

The series will be in the form of a deep dive into the code with explanations and walkthroughs along side.

Next up will be a series of Posts that will take you through a few concepts in the most beginner appealing way that we can put up.

So What exactly is Reinforcement Learning?

You will receive a full Mathematically and Programmatically sound answer in the series, but here is a fun one to begin with.

Imagine you’re in a Bakery and have been told to bake a Delicious cake by your Supervisor.

Your Supervisor is rather a strict person who leaves you to discover the best recipe. However, since the Supervisor hates you, she will thrash you every time you bake a bad cake (Probably not the best place to work at).

Now, you’re a smart kid! You start out an experiment. You keep a track of your Performance and the taste of every attempt.

Your end goal is to impress your Supervisor (maximise your reward). You start out as an inexperienced person. You play around the Bakery (Your environment) and keep trying until you finally impress your Supervisor (Reward)

You start out by adding Salt, by burning down a few things and get Thrashed every time you do so (Receive a penalty) and since you’re smart, you make sure you don’t do this again (Keep a track of previous moves).

In the end you finally get ‘Trained’ once you’ve baked the Best Cake and received your highest goal.

So this is how RL works.

  • There is an Agent: You.
  • In an Environment: Bakery.
  • Whose goal is to maximise the Reward: Receive an appraisal from the Supervisor.
  • The agent keeps getting feedback from the Environment: The Supervisor.
  • A positive Feedback for a right move and a Negative feedback for a wrong step.
  • The Agent has a Memory of it’s previous actions, based on which it Learns.
  • The Agent keeps interacting until the Reward is Maximised.
This post has been created by Sanyam Bhutani and reviewed by Lucas Vazquez
Lucas and the team init 27 are also devoloping a RL Library, we really need collaborators to our Library. Ping us here you’re interested, we’d love to have you on board!
Subscribe to our Newsletter for a Weekly Curated List of Deep Learning and Computer Vision
One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.