Abduction in machine learning

Both inductive learning and abductive reasoning start from specific facts or observations and produce some explanation of these facts. Both may be described as forms of defeasible reasoning from effects to causes. There are some differences, but they are minor and due to different understandings of the notions of observation and explanation (see for instance [Bergadano and Besnark, 1994]). We build on the general notions developed in the introductory Chapter, taking what was labeled there as the syllogistic view, in the sense that we isolate the differences between abduction and induction based on syntactic considerations.