Part 1 : Evolutionary Algorithms for Reinforcement Learning — An Introduction

An Introduction to Genetic algorithm

Unlike conventional gradient descent based Reinforcement learning techniques, there are alternative optimization techniques which are inspired by Charles Darwin’s Evolutionary theory. In this Article we deep dive to understand the working of these algorithms and try to implement them to solve OpenAI’s cartpole problem

Neural Networks, the building blocks of the modern AI are inspired by biological brain. Similarly convolution neural networks are inspired by biological vision. Nature through the process of evolution has solved many problems efficiently and elegantly. Hence, many advances in technology and engineering are inspired by biological organisms. Naturally people have tried to borrow the process of evolution itself and implement it as a computer algorithm to solve problems. In this article we will see how process of evolution is harnessed to solve problems.

Why different approach to Reinforcement Learning?

With DQN and policy gradient approaches we create a agent whose policy depended on a neural network to approximate the Q function or policy function. The agent interacts with the environment, collects experiences and then backpropagates to improve the accuracy of the…