Reinforcement learning is a method of engineering intelligence that emulates biological organisms by transducing information from the environment, processing it, and outputting behavior conducive to survival. It borrows from behaviorism and cognitive science to model agent-environment interactions. Dynamic programming is a mathematical optimization method used in reinforcement learning, but it has limitations that are addressed by model-free approaches and the use of artificial neural networks. The combination of reinforcement learning and artificial neural networks, such as Deep Q-Learning, shows promise in improving the capabilities of AI. However, there are still challenges in achieving artificial general intelligence, as it requires a complex internal architecture and reflective awareness.
Table of contents
A deep dive into the rudiments of reinforcement learning, including model-based and model-free methodsWhat is Reinforcement Learning?Decision Theory & Control TheoryStates, Actions & RewardsQuantifying RewardMarkov Decision Process (MDP)Dynamic Programming & Bellman OptimalityState-Value FunctionAction-Value FunctionModel Free Methods: Monte Carlo & Temporal DifferenceAugmenting Reinforcement Learning with ANNsOff-Policy DQNOn-Policy Deep TD(𝝀)Reinforcement Learning and Artificial General IntelligenceSelected ReferencesSort: