How to build a Reinforcement-Learning model in AnyLogic

Nov 9 / Benjamin Schumann
“Can you please share how you created that model?” “I would love to see how one you combine machine-learning and simulation in practice”…

I got many such messages after a recent post where I pondered about how simulation and machine-learning/AI can collaborate better. 

So let's do it together, from scratch...

Series overview

Hence, I decided to try a new format: in the next few weeks, I will publish a four-part video series exploring a conceptual AnyLogic model that applies reinforcement learning. It is build from scratch, i.e. doesn’t apply any external libraries and black-box approaches. This is probably the best way to learn about it :-)

Please note that all rights are reserved, so do not use this without my permission on anything but learning :-) 

The model

You can always run and play with the model yourself by clicking below. If it doesn’t work, you can play it directly on the AnyLogic cloud here.

Part 1 - introduction

You can view the first part below or directly on YouTube
Write your awesome label here.

Part 2 - actual agent structure

In this second part, I will introduce the model in more detail and we will dive deeper into the actual agents. See below or directly here.
Write your awesome label here.

Part 3 - algorithms

This week, we will examine the flesh and bones of the model, looking into the actual agent classes used and some of their algorithms. See below or directly here.
Write your awesome label here.

Part 4 - the juicy bits

In this fourth and last part, we go even deeper into model, exploring how the agents are created and how the algorithms make it all work. We focus on the actual Bellman equation and how it creates the actual learning effect. See below or directly here
Write your awesome label here.
Created with