Semantics of probabilistic programs

Abstract

This paper presents two complementary but equivalent semantics for a high level probabilistic programming language. One of these interprets programs as partial measurable functions on a measurable space. The other interprets programs as continuous linear operators on a Banach space of measures. It is shown how the ordered domains of Scott and others are embedded naturally into these spaces. We use the semantics to prove a general result about probabilistic programs, namely, that a program's behavior is completely determined by its action on fixed inputs. © 1981.