1 of 23

Deep Reinforcement Learning �- 1. Course Overview

1

2 of 23

Outline

  1. Syllabus
  2. Basic of Reinforcement Learning

2

3 of 23

Syllabus

3

4 of 23

Syllabus

4

5 of 23

Syllabus

5

6 of 23

Syllabus

6

7 of 23

Resources

7

8 of 23

1. Basic of Reinforcement Learning

8

9 of 23

1. Basic of Reinforcement Learning

  • Reinforcement Learning:�

9

10 of 23

1. Basic of Reinforcement Learning

  • Reinforcement Learning:�

Supervised �Learning���

Unsupervised�Learning

ReinforcementLearning

Semi-sv

10

11 of 23

1. Basic of Reinforcement Learning

11

12 of 23

1. Basic of Reinforcement Learning

  • Reinforcement Learning:�At each step, the agent: �(1) Observe new state�(2) Executes action�(3) Receive reward

1. State (observe)

2. Action

3. Reward

I have a Goal!!�Ex: Let’s save the world

Environment � (world)

Agent�(model)

12

13 of 23

1. Basic of Reinforcement Learning

  • The Challenge for RL in Real-World Applications:� Environment should give reward in real time�Open Challenges.

13

14 of 23

1. Basic of Reinforcement Learning

  • Reinforcement Learning:�Input: Given environment which provides numerical reward signal, and agent which act inside of that environment�Outputs: Let agent learn how to take actions(policy) in order to maximize reward.

���

  • Goal: Learn how to take actions in order to maximize reward
  • Design RL: Objective, State, Action, Reward

14

15 of 23

1. Basic of Reinforcement Learning

  • Design RL: Objective, State, Action, Reward

15

16 of 23

1. Basic of Reinforcement Learning

  • Design RL: Objective, State, Action, Reward

16

17 of 23

1. Basic of Reinforcement Learning

  • Design RL: Objective, State, Action, Reward

17

18 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Robot in Room Example

18

19 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Robot in Room Example

19

20 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Robot in Room Example

20

21 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Robot in Room Example

21

22 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Robot in Room Example

Lessons from Robot in Room

  • Environment model has big impact on optimal policy
  • Reward structure has big impact on optimal policy

22

23 of 23

1. Basic of Reinforcement LearningHow to design good RL problem formulation? �- Course runner Example

Reward Structure may have Unintended Consequences

Player gets reward based on:

  1. Finishing time
  2. Finishing position
  3. Picking up “turbos”

23