Logical Generative Models for Probabilistic Reasoning about Existence, Roles and Identity

In probabilistic reasoning, the problems of existence and identity are important to many different queries; for example, the probability that something that fits some description exists, the probability that some description refers to an object you know about or to a new object, or the probability that an object fulfils some role. Many interesting queries reduce to reasoning about the role of objects. Being able to talk about the existence of parts and sub-parts and the relationships between these parts, allows for probability distributions over complex descriptions. Rather than trying to define a new language, this paper shows how the integration of multiple objects, ontologies and roles can be achieved cleanly. This solves two main problems: reasoning about existence and identity while preserving the clarity principle that specifies that probabilities must be over well defined propositions, and the correspondence problem that means that we don’t need to search over all possib...

• Type: Conference
• Year: 2007
• Where: AAAI
• Authors: David Poole
• Keywords: AAAI 2007, Intelligent Agents, Many Interesting Queries, Probabilistic Reasoning, Probability Distributions