Modular and taxonomic representations for tractable reasoning about actions

点击翻译按钮获取中文摘要

A long standing and fundamental problem of AI research is this: the development of a comprehensive and general theory for representation and efficient reasoning about actions of agents operating in uncertain and incompletely known large-scale domains. The proposed research will contribute towards this long-term goal by eliciting structured logical representations and subsequently elaborating them to guarantee efficient reasoning in structured domains. The proposed research focuses on representations of actions formulated in order-sorted first order logic and in description logic-based languages. The primary objective of the proposed research program is the theoretical development of those representations of actions and their effects that can be characterized as modular, taxonomic or structural and the experimental assessment of these representations on solving the common reasoning tasks. This research is original and innovative for several reasons. First, it emphasizes the development of logical theories of actions that exploit the declarative structure of application domains: taxonomies of actions and objects. Second, it proposes to integrate taxonomic reasoning about actions and about objects with order-sorted logics and modular description logic ontologies. Finally, it proposes to use these representational leverages to solving challenging and important problems in AI, including decision-theoretic planning and learning. While the proposed research program will help to advance our knowledge, it also has a significant promise of developing new logical frameworks, information technologies (and subsequently, software products) that will be relevant for a broad range of applications including health-care, Semantic Web services and industry.