An adaptive logic based approach to abduction in AI

TitleAn adaptive logic based approach to abduction in AI
Publication TypeConference Proceedings
Year of Publication2011
AuthorsGauderis, T
Conference NameNinth International Workshop on Non-Monotonic Reasoning, Action and Change

In a logic-based approach to abductive reasoning, the background knowledge is represented by a logical theory. A sentence &\#934; is then considered as an explanation for ω if it satisfies some formal conditions. In general, the following three conditions are considered crucial: (1) Φ together with the background knowledge implies !; (2) Φ is logically consistent with what is known; and (3) Φ is the most ‘parsimonious’ explanation. But, since abductive reasoning is a non-monotonic form of reasoning, each time the background knowledge is extended, the status of previously abduced explanations becomes once again undefined. The adaptive logics program is developed to address these types of non-monotonic reasoning. In addition to deductive reasoning steps, it allows for direct implementation of defeasible reasoning steps, but it adds to each formula the explicit set of conditions that would defeat this formula. So, in an adaptive logic for abduction, a formula is an abduced hypothesis as long as none of its conditions is deduced. This implies that we will not have to recheck all hypotheses each time an extension to our background knowledge is made. This is the key advantage of this approach, which allows us to save repetitive re-computations in fast growing knowledge bases.

Citation Key3178952
Download PDF (Author PDF)
PDF author (public):