Learning Faster by Discovering and Exploiting Object Similarities

[摘要] In this paper we explore the question: “Is it possible to speed up the learning process of an autonomous agent by performing experiments in a more complex environment (i.e., an environment with a greater number of different objects)?” To this end, we use a simple robotic domain, where the robot has to learn a qualitative model predicting the change in the robot's distance to an object. To quantify the environment's complexity, we defined cardinal complexity as the number of objects in the robot's world, and behavioural complexity as the number of objects' distinct behaviours. We propose Error reduction merging (ERM), a new learning method that automatically discovers similarities in the structure of the agent's environment. ERM identifies different types of objects solely from the data measured and merges the observations of objects that behave in the same or similar way in order to speed up the agent's learning. We performed a series of experiments in worlds of increasing complexity. The results in our s...

[学科分类] 自动化工程

[关键词] Autonomous Learning Agents;Learning Speed;Domain Complexity;Learning by Experimentation;Machine Learning