5.3 Learning Technologies - Q-Learning

Learning Technologies - Q-Learning

In a nutshell, Q-Learner is basically what the AI has learnt in artificial neural networks (5.2 Learning Technologies - Artificial Neural Networks) and applying them to an in game situation.

"The Q-learning algorithm is a reinforcement learning algorithm. Reinforcement learning algorithms are a set of machine learning algorithms inspired by behavioral psychology. The basic premise is that you teach the algorithm to take certain actions based on prior experience by rewarding or punishing actions. Similar to teaching a dog to sit by giving it treats for good behavior." (20)

Now what Soren is saying is that Q-Learning is a learning algorithm, well at the end of the day its used to teach the AI to take on actions based on is past experience. Soren refers to the AI as a dog which is being reward or punished based on its actions, In somewhat I suppose his theory is correct but I like to think to Q-Learning as a real human, the stages of life, a baby is born and is knew to everything, then as it gets older it gains experience and knowledge and learners from its mistakes, if it does well it is rewarded as so forth. Q-Learning is so linked to human life and ANN's that it combines the human way of life and apply's that to a gaming AI. 

I have talked about how Q-Learning relates to artificial neural networks in (5.2 Learning Technologies - Artificial Neural Networks - If your playing a 1 v 1 match, player character vs the AI. If you were to stay in the same location\hiding from the AI and manage to kill the AI character in your current location, because the AI character did not know your whereabouts, when the AI re-spawns, from its experience and using artificial neural networks the AI will have learned and will remember where the player character is hiding and wont make the same mistake again by running into your path.")

Two previous algorithms come together to create Q-Learning, these are the Bellman equation: which describes the value of a problem at a certain point in time made from previous decisions, its used by breaking a problem down into smaller problems. The other algorithm is The Law of Effect: which is the process of trial and error. By combining the two algorithms, you get Q-Learning. A perfect example of gaming AI using academic Q-learning is Forza on Xbox One, you can create a drivatar which is an AI which learns how you drive in the game by imitating and repeating your driving style.