Mario Román

Home

❯

notes

❯

Substochastic versus unnormalized kernels

Substochastic versus unnormalized kernels

Jan 16, 20261 min read

Roughly, the following works prefers unnormalized kernels over substochastic kernels.

  • Commutative Semantics for Probabilistic Programming (Staton, 2017) uses unnormalized kernels from the abstract.
  • Semantics of Probabilistic Programs (Kozen, 1981) uses cones, like positive subsets of powers of the reals.
  • Contextual Equivalence for Probabilistic Programs with Continuous Random Variables and Scoring (Culpepper, Cobb) explicitly argues against subdistributions.
  • An Introduction to Probabilistic Programming (van de Meent, Paige, Yang, Wood, 2021)
  • Measurable Cones and Stable, Measurable Functions (Ehrhard, Pagani, Tasson, 2017)

The following is literature that prefers substochastic kernels over unnormalized kernels.

  • The Category of Markov Kernels (Panangaden, 1999), substochastic Giry monad.
  • Labelled Markov Processes (Panangaden), substochastic Giry monad.
  • A Lambda-Calculus Foundation for Universal Probabilistic Programming (Borgstrom, Dal Lago, et al)
  • Lambda Calculus and Probabilistic Computation (Faggian, Ronchi della Rocca, 2019)
  • Partial Markov Categories (Di Lavore, Roman, Sobocinski, 2025), subprobability Kleisli categories.

Graph View

Mario Román (2026), CC-BY-SA. Built mostly with Quartz and Write.

  • GitHub
  • ArXiv
  • OrcID