RankPL: a qualitative probabilistic programming language[go to overview]
In recent years, probabilistic programming languages (PPLs) have become a popular tool in the field of Bayesian Machine Learning. Roughly speaking, PPLs allow one to specify and reason about probabilistic models, by writing programs that include probabilistic random choice constructs and observation statements.
In this talk I introduce a qualitative variant of a PPL called RankPL. RankPL can be used to reason about uncertainty expressible by distinguishing normal from exceptional events. This kind of uncertainty often appears in commonsense reasoning problems, where precise probabilities are unknown. Semantically, RankPL is based on a qualitative abstraction of probability theory called ranking theory.
I discuss syntax and semantics of RankPL, its relation to (iterated) belief revision, and I present a number of motivating examples of RankPL programs.
16.11.17 - 10:15