Deep reinforcement learning
Deep reinforcement learning (deep RL) is a subfield of RL that combines RL with deep learning. In other words, the idea behind it is to exploit the learning capabilities of a neural network to solve RL problems. In traditional RL, policies and value functions are represented by simple functions. These methods work well with low-dimensional state and action spaces (i.e., when the environment and agent can be easily modeled). When the environment becomes more complex or larger, traditional methods fail to generalize. In deep RL, instead, policies and value functions are represented by neural networks. A neural network can theoretically represent any complex function (Universal Approximation Theorem), and this allows deep RL methods to solve problems with high-dimensional state spaces (such as those presenting images, videos, or continuous tasks). Modeling complex functions thus allows the agent to learn a more generalized and flexible policy that is needed...
