CptS 580-03: Intelligent Agents
Spring 2014

Matthew E. Taylor (Matt)
EME 137
Syllabus: Spring 2014
Reinforcement Learning: An Introduction  


Reading Response
1/14 First day of class: Introduction  
1/16 Secord day of class: Introduction, continued Chapter 1
1/21 Bandits Chapter 2
1/23 Reinforcement Learning Problem Definition Chapter 3
1/28 Dynamic Programming Chapter 4
1/30 Dynamic Programming  
2/4 Monte Carlo Methods Chapter 5
2/6 Monte Carlo Methods, Temporal Difference Learning Chapter 6: 6.1-6.3
2/11 Temporal Difference Learning Chapter 6: 6.4-6.10
Sign up for a time to present to the class: http://doodle.com/2p2wz39vebwdzktn
2/13 Eligibility Traces Chapter 7
2/18 Generalization and Function Approximation Chapter 8
2/20 Generalization and Function Approximation IFSA: Incremental Feature-Set Augmentation for Reinforcement Learning Tasks
2/25 Planning and Learning Chapter 9
2/27 Planning and Learning Read the R-Max paper, at least through section 3.
3/4 IRL, Yusen Presentation/Discussion on Inverse Reinforcement Learning
Matt's later slides
3/6 Multi-agent RL, Xiyu Presentation/Discussion, Matt's later slides  
3/11 David Presentation/Discussion on Robotics, Matt's notes  
3/13 Anthony Presentation/Discussion on Multiagent RL  
3/25 Bei Presentation/Discussion on Learning from Human Rewards  
3/27 Chris Presentation/Discussion on RL in traffic control, Matt's notes  
4/1 Beiyu Presentation/Discussion on hierarchical RL Final project proposal due by 3/31 at 3:00pm. Please submit via Angel
4/3 Dmitry Presentation/Discussion on RL in SLAM
Two Videos
Matt's Notes
4/8 Gabe Presentation/Discussion on learning in quadcopters  
4/10 Josh Presentation/Discussion on Game Playing
Discussion on Function Approximation (on board)
4/15 Transfer Learning 4th Exercise due
Read Transfer in Reinforcement Learning: a Framework and a Survey by A. Lazaric and write a response. Read sections 1, 2, 6. Also, read one of section 3, 4, or 5.
Please vote on what's next in the class.
4/17 Transfer Learning, continued  
4/22 POMDPs Read the tutorial here.
At a minimum, read from "Background on POMDPs" through "General Form of a POMDP solution."
No reading response is required.
4/24 Reward Shaping  
4/29 Intrinsic RL  
5/1 5 min presentations on final project progress  
5/2Last day to ask Matt questions before he leaves the country 
5/8   Final project due on Angel by 11:59pm

Possible further topics
  • Current Function Approximation Choices
  • Efficient Model-Learning methods
  • Hierarchical Methods
  • Game Playing
  • Learning in Robotics
  • Transfer Learning
  • Shaping Rewards
  • Learning from Human Rewards
  • Learning from Demonstration
  • Multiagent RL
  • Partially observable envirnments and/or POMDPs
  • Meta-RL and empirical evaluation of algorithms
  • Least Squares methods (e.g., LSPI)
  • Adaptive Representations / Representation Learning
  • Case Studies: Robot soccer, Helicopter Control, etc.
  • Inverse Reinforcement Learning (IRL)
  • Intrinsicly Motivated Reinforcement Learning
  • Actor-Critic Methods
  • Policy Gradient methods
  • Crowd Sourcing (?)