No abstract

We make a formal analogy between random sampling and fresh name generation. We show that quasi-Borel spaces, a model for probabilistic programming, can soundly interpret the ν-calculus, a calculus for name generation. Moreover, we prove that this semantics is fully abstract up to first-order types. This is surprising for an ‘off-the-shelf’ model, and requires a novel analysis of probability distributions on function spaces. Our tools are diverse and include descriptive set theory and normal forms for the ν-calculus.

We define a probabilistic programming language for Gaussian random variables with a first-class exact conditioning construct. We give operational, denotational and equational semantics for this language, establishing convenient properties like exchangeability of conditions. Conditioning on equality of continuous random variables is nontrivial, as the exact observation may have probability zero; this is Borel's paradox. Using categorical formulations of conditional probability, we show that the good properties of our language are not particular to Gaussians, but can be derived from universal properties, thus generalizing to wider settings. We define the Cond construction, which internalizes conditioning as a morphism, providing general compositional semantics for probabilistic programming with exact conditioning.

We introduce extended Gaussian distributions as a precise and principled way of combining Gaussian probability uninformative priors, which indicate complete absence of information. To give an extended Gaussian distribution on a finite-dimensional vector space X is to give a subspace D, along which no information is known, together with a Gaussian distribution on the quotient X/D. We show that the class of extended Gaussians remains closed under taking conditional distributions. We then introduce decorated linear maps and relations as a general framework to combine probability with nondeterminism on vector spaces, which includes extended Gaussians as a special case. This enables us to apply methods from categorical logic to probability, and make connections to the semantics of probabilistic programs with exact conditioning.