Description
Springer Design of Experiments for Reinforcement Learning by Christopher Gatti
This thesis takes an empirical approach to understanding of the behavior and interactions between the two main components of reinforcement learning: the learning algorithm and the functional representation of learned knowledge. The author approaches these entities using design of experiments not commonly employed to study machine learning methods. The results outlined in this work provide insight as to what enables and what has an effect on successful reinforcement learning implementations so that this learning method can be applied to more challenging problems._x000D_ Table of contents : - _x000D_
GLOSSARY_x000D_
ACKNOWLEDGMENT_x000D_
FOREWARD_x000D_
1. INTRODUCTION _x000D_
2. REINFORCEMENT LEARNING_x000D_
2.1 Applications of reinforcement learning _x000D_
2.1.1 Benchmark problems_x000D_
2.1.2 Games_x000D_
2.1.3 Real-world applications_x000D_
2.1.4 Generalized domains_x000D_
2.2 Components of reinforcement learning_x000D_
2.2.1 Domains_x000D_
2.2.2 Representations_x000D_
2.2.3 Learning algorithms _x000D_
2.3 Heuristics and performance effectors _x000D_
3. DESIGN OF EXPERIMENTS _x000D_
3.1 Classical design of experiments_x000D_
3.2 Contemporary design of experiments_x000D_
3.3 Design of experiments for empirical algorithm analysis _x000D_
4. METHODOLOGY _x000D_
4.1 Sequential CART_x000D_
4.1.1 CART modeling _x000D_
4.1.2 Sequential CART modeling_x000D_
4.1.3 Analysis of sequential CART _x000D_
4.1.4 Empirical convergence criteria _x000D_
4.1.5 Example: 2-D 6-hump camelback function_x000D_
4.2 Kriging metamodeling _x000D_
4.2.1 Kriging _x000D_
4.2.2 Deterministic kriging_x000D_
4.2.3 Stochastic kriging _x000D_
4.2.4 Covariance function _x000D_
4.2.5 Implementation _x000D_
4.2.6 Analysis of kriging metamodels_x000D_
5. THE MOUNTAIN CAR PROBLEM _x000D_
5.1 Reinforcement learning implementation _x000D_
5.2 Sequential CART_x000D_
5.3 Response surface metamodeling_x000D_
5.4 Discussion _x000D_
6. THE TRUCK BACKER-UPPER PROBLEM_x000D_
6.1 Reinforcement learning implementation_x000D_
6.2 Sequential CART _x000D_
6.3 Response surface metamodeling_x000D_
6.4 Discussion _x000D_
7. THE TANDEM TRUCK BACKER-UPPER PROBLEM _x000D_
7.1 Reinforcement learning implementation_x000D_
7.2 Sequential CART _x000D_
7.3 Discussion _x000D_
8. DISCUSSION _x000D_
8.1 Reinforcement learning_x000D_
8.2 Experimentation _x000D_
8.3 Innovations _x000D_
8.4 Future work _x000D_
APPENDICES_x000D_
A. Parameter effects in the game of Chung Toi_x000D_
B. Design of experiments for the mountain car problem _x000D_
C. Supporting tables_x000D_