AcademIc Coffee: Deep Q-Learning
Reinforcement learning is an area in machine learning which is focused mainly on training an agent in a given environment in order to get the maximum reward based on the action it can take at the moment.
Q-Learning refers to a technique for reinforcement learning in which we hypothesize a function for an agent which knows nothing about the environment except for the current state and possible actions it can take. This function is called the Q function which is a reference for the ‘Quality’ of the decision made, the traditional Q-function value can be obtained with help of the bellman equation and its explained more in depth in the post in the link below.
This function helps out agent to make the decision which maximizes our reward on the long term, since we take into consideration the future decisions made.
In Deep Q-Learning we make all these assumptions, except that we train a model in order to predict for the value of the Q-function which helps us make the best decision.
This technique has been used to maximize rewards in different environments such as basic games or stock market simulations.
More details can be found in the following post:
