NURTURING PROMOTES THE EVOLUTION OF LEARNING IN CHANGING ENVIRONMENTS

摘要

An agent may interact with its environment and learn complex tasks based on evaluative
\nfeedback through a process known as reinforcement learning. Reinforcement
\nlearning requires exploration of unfamiliar situations, which necessarily involves unknown
\nand potentially dangerous or costly outcomes. Supervising agents in these
\nsituations can be seen as a type of nurturing and requires an investment of time usually
\nby humans. Nurturing, one individual investing in the development of another
\nindividual with which it has an ongoing relationship, is widely seen in the biological
\nworld, often with parents nurturing their o spring. There are many types of nurturing,
\nincluding helping an individual to carry out a task by doing part of the task for
\nit. In arti cial intelligence, nurturing can be seen as an opportunity to develop both
\nbetter machine learning algorithms and robots that assist or supervise other robots.
\nAlthough the area of nurturing robotics is at a very early stage, the hope is that this
\napproach can result in more sophisticated learning systems. This dissertation demonstrates
\nthe e ectiveness of nurturing through experiments involving the evolution of
\nthe parameters of a reinforcement learning algorithm that is capable of nding good
\npolicies in a changing environment in which the agent must learn an episodic task
\nin which there is discrete input with perceptual aliasing, continuous output, and delayed
\nreward. The results show that nurturing is capable of promoting the evolution
\nof learning in such environments.