Tradeoffs in Learning and Planning

这项研究的总体目标是使之有可能 构建机器学习系统， 在复杂的现实世界中的规划性能 域. 目前的机器学习技术并不 因为他们的语言， 表示的表达能力不足以捕捉知识 必须在这些领域发挥作用。 这项工作将研究 语言的代表性之间的关系 学习的知识和涉及的计算权衡 用这种语言学习和规划。 我们亦建议 通过开发，提高现有技术的能力 学习和规划方法， 表示。

The overall objective of this research is to make it possible to build machine learning systems that robustly improve their planning performance with experience in complex real-world domains. The current machine learning techniques are not adequate for this purpose because their language of representation is not expressive enough to capture the knowledge needed to be effective in these domains. This work will study the relationship between the language of representation of the learned knowledge and the computational tradeoffs involved in learning and planning in that language. We also propose to enhance the capabilities of current techniques by developing learning and planning methods for more expressive representations.