Reasoning with imperfect information

Publisher Summary Handling imperfect information is a crucial problem for any intelligent system that has to deal with the real world. Many proposals have been put forward to solve this problem. This chapter investigates some of the most important of these proposals and, discusses their advantages, limitations, and interrelationships. It also provides some examples of their relevance in building intelligent systems. This chapter focuses on the three main numerical methods: probability theory, evidence theory, and possibility theory. It also discusses three main symbolic methods: circumscription, default logic, and autoepistemic logic. The chapter contains descriptions of some of the more intriguing lesser models. This chapter suggests reasons for the introduction of the more novel techniques and lays out the technical differences between the approaches. The majority of the techniques suggested to handle information that is certain but incomplete have been symbolic—often being variations of first order or prepositional logic.Such techniques involve making assumptions to complete what is known. The bulk of the techniques for handling complete but uncertain information are basically numerical. This chapter also describes the relationships between techniques and within their genres. It closes with a discussion on applications of these techniques.