1 / 21

Reinforcement Learning

Reinforcement Learning. Michael Roberts. With Material From: Reinforcement Learning: An Introduction Sutton & Barto (1998). What is RL?. Trial & error learning without model with model Structure. s3. r2. r1. s1. s2. r3. s4. RL vs. Supervised Learning.

oriel
Télécharger la présentation

Reinforcement Learning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Reinforcement Learning Michael Roberts With Material From: Reinforcement Learning: An Introduction Sutton & Barto (1998)

  2. What is RL? • Trial & error learning • without model • with model • Structure s3 r2 r1 s1 s2 r3 s4

  3. RL vs. Supervised Learning • Evaluative vs. Instructional feedback • Role of exploration • On-line performance

  4. K-armed Bandit Problem Average Rewards Actions 10 0, 0, 5, 10, 35 5, 10, -15, -15, -10 -5 Agent 100 0

  5. K-armed Bandit Cont. • Greedy exploration • ε-greedy • Softmax Average Reward: Incremental formula: where: α = 1 / (k+1) Probability of choosing action a:

  6. More General Problems • More than one state • Delayed rewards • Markov Decision Process (MDP) • Set of states • Set of actions • Reward function • State transition function • Table or Function Approximation

  7. Example: Recycling Robot

  8. Recycling Robot: Transition Graph

  9. Dynamic Programming

  10. Backup Diagram .25 .25 .25 .4 .6 .7 .3 .5 .5 Rewards 10 5 200 200 -10 1000

  11. Dynamic Programming:Optimal Policy

  12. Backup for Optimal Policy

  13. Performance Metrics • Eventual convergence to optimality • Speed of convergence to optimality • Regret (Kaelbling, L., Littman, M., & Moore, A. 1996)

  14. Gridworld Example

  15. Initialize V arbitrarily, e.g.          , for all         Repeat For each until        (a small positive number) Output a deterministic policy, such that:

  16. Temporal Difference Learning • RL without a model • Issue of: temporal credit assignment • Bootstraps like DP • TD(0):

  17. TD Learning • Again, TD(0) = TD(λ) = where e is called an eligibility trace

  18. Backup Diagram for TD(λ)

  19. TD-Gammon (Tesauro)

  20. Additional Work • POMDP’s • Macros • Multi-agent rl • Multiple reward structures

More Related