Combining Genetic Algorithms with Memory Based Reasoning

摘要

Combining different machine learning algorithms in the same system can produce benefits above and beyond what either method could achieve alone. This paper demonstrates that genetic algorithms can be used in conjunction with memory-based reasoning to solve a difficult class of delayed reinforcement learning problems that both methods have trouble solving individually. This class includes numerous important control problems that arise in robotics, planning, game playing, and other areas. Our experiments demonstrate that by using one learning technique, genetic algorithms, as a bootstrapping method for the second learning technique, memory-based reasoning, we can create a system that outperforms either method alone. The resulting joint system learns to solve a difficult reinforcement learning task with a high degree of accuracy and with relatively small memory requirements. 1 INTRODUCTION When two people learn a task together, they can both benefit from the different skills that each br...