# Deep Reinforcement Learning ### Intro from CS 285 RL is fundamentally two things: 1. A mathematical formalism for learning-based decision making (this allows us to design algorithms) 2. An approach for learning decision making and control **from experience**(rather than relying on policy being designed by hand) How is this different from other ML topics?  In RL, we model a decision making system as an interaction between an agent and an environment. The **agent** makes **decisions (actions)** and the **environment** responds to those actions with **observations** (which we call **states**) and **rewards**. Crucially, this process is repeated multiple times. The real world is unstructured and unpredictable. [Deep-Learning](Deep-Learning.md) helps us handle *unstructured environments*. In DL we train a very large, heavily over parameterized model to map inputs to outputs. RL provides a formalism for *behavior*. --- Date: 20211124 Links to: Tags: References: * [Great intro course, combined with DL](https://www.youtube.com/watch?v=JHrlF10v2Og&list=PL_iWQOsE6TfURIIhCrlt-wj9ByIVpbfGc)