### 你将学到什么

Formalize problems as Markov Decision Processes

Understand basic exploration methods and the exploration / exploitation tradeoff

Understand value functions, as a general-purpose tool for optimal decision-making

Know how to implement dynamic programming as an efficient solution approach to an industrial control problem

### 课程概况

Reinforcement Learning is a subfield of Machine Learning, but is also a general purpose formalism for automated decision-making and AI. This course introduces you to statistical learning techniques where an agent explicitly takes actions and interacts with the world. Understanding the importance and challenges of learning agents that make decisions is of vital importance today, with more and more companies interested in interactive agents and intelligent decision-making.

This course introduces you to the fundamentals of Reinforcement Learning. When you finish this course, you will:

– Formalize problems as Markov Decision Processes

– Understand basic exploration methods and the exploration/exploitation tradeoff

– Understand value functions, as a general-purpose tool for optimal decision-making

– Know how to implement dynamic programming as an efficient solution approach to an industrial control problem

This course teaches you the key concepts of Reinforcement Learning, underlying classic and modern algorithms in RL. After completing this course, you will be able to start using RL for real problems, where you have or can specify the MDP.

This is the first course of the Reinforcement Learning Specialization.

### 课程大纲

**Welcome to the Course! **

Welcome to: Fundamentals of Reinforcement Learning, the first course in a four-part specialization on Reinforcement Learning brought to you by the University of Alberta, Onlea, and Coursera. In this pre-course module, you'll be introduced to your instructors, get a flavour of what the course has in store for you, and be given an in-depth roadmap to help make your journey through this specialization as smooth as possible.

**The K-Armed Bandit Problem**

For the first week of this course, you will learn how to understand the exploration-exploitation trade-off in sequential decision-making, implement incremental algorithms for estimating action-values, and compare the strengths and weaknesses to different algorithms for exploration. For this week’s graded assessment, you will implement and test an epsilon-greedy agent.

**Markov Decision Processes**

When you’re presented with a problem in industry, the first and most important step is to translate that problem into a Markov Decision Process (MDP). The quality of your solution depends heavily on how well you do this translation. This week, you will learn the definition of MDPs, you will understand goal-directed behavior and how this can be obtained from maximizing scalar rewards, and you will also understand the difference between episodic and continuing tasks. For this week’s graded assessment, you will create three example tasks of your own that fit into the MDP framework.

**Value Functions & Bellman Equations **

Once the problem is formulated as an MDP, finding the optimal policy is more efficient when using value functions. This week, you will learn the definition of policies and value functions, as well as Bellman equations, which is the key technology that all of our algorithms will use.

**Dynamic Programming**

This week, you will learn how to compute value functions and optimal policies, assuming you have the MDP model. You will implement dynamic programming to compute value functions and optimal policies and understand the utility of dynamic programming for industrial applications and problems. Further, you will learn about Generalized Policy Iteration as a common template for constructing algorithms that maximize reward. For this week’s graded assessment, you will implement an efficient dynamic programming agent in a simulated industrial control problem.