Reinforcement Learning for Video Games

Abstract

Reinforcement Learning is a way of machine learning in which the training of machine learning models takes place by training the machine (the agent) to choose a certain action from a different set of actions in a given state from a different set of states in a given environment and by taking that action the agent receives a reward which will help it choose better actions in the future by maximizing this reward over time. Reinforcement learning is making the agent learn by interacting with the environment without the need to collect samples to train on, in traditional reinforcement learning there is usually a Q-Table that is basically a table that consists of all possible actions as columns and all possible states as rows, each cell (pair of action-state) has a value which is called Q-Value which is the reward for taking that action at that state in the given environment, However traditional Reinforcement learning is not very efficient in the modern complex environments so comes the need of the scope of this project: Deep Reinforcement learning.