Dependent Bayesian Lenses: Categories of Bidirectional Markov Kernels with Canonical Bayesian Inversion — Dylan Braithwaite & Jules Hedges (2022). arXiv:2209.14728 (v1, PDF).
Constructs categories of bidirectional Markov kernels in which Bayesian inversion is canonical: by passing to supports of states, inverses become unique rather than unique almost surely, and inversion becomes a genuine functor into a category of dependent (fibred) Bayesian lenses.
Sources: the paper, arXiv:2209.14728v1, checked against the arXiv listing. Index: Papers.
Key definitions and results
- Definitions 2–3: almost-sure equality, Bayesian inversion
- Definitions 4, 6: supports and Bayesian inverses with support
- Proposition 7, Corollary 8: uniqueness