Paolo Perrone from the University of Oxford visited the Paris office and was kind enough to give a talk about his work on categories for probability theory.
The recording reveals that Arnaud Spiwack's computer was used to stream voice, but not the camera feed. So you don't get to see Paolo. Maybe it's for the best: we were having camera issues.
Abstract:
Markov categories are one of the most recent abstract frameworks for probability, statistics, and related fields.
They can be considered a higher abstraction layer on top of measure theory, where one can work with concepts such as disintegrations and regular conditionals without using measure theory directly.
In the past few years they have been successfully employed to restate, reprove, and even generalize core concepts of probability theory, from de Finetti's theorem to d-separation criteria, only using diagrammatic manipulations.
In this talk I will explain the main theory of Markov categories and their relationship with monads, with particular emphasis on conditioning and Bayesian inversions.
Given time and interest, I can also show how to apply these ideas to statistical experiments, and extend Blackwell's theorem, for the first time, beyond the discrete case.
Relevant papers:
- P. Perrone, "Markov Categories and Entropy", IEEE Transactions of Information Theory 70(3), 2024. (arXiv:2212.11719)
- T. Fritz, T. Gonda, P. Perrone and E. F. Rischel, "Representable Markov categories and comparison of statistical experiments in categorical probability", Theoretical Computer Science 961, 2023. (arXiv:2010.07416)
- N. Ensarguet and P. Perrone, "Categorical probability spaces, ergodic decompositions, and transitions to equilibrium", submitted. (arXiv:2310.04257)
--------------------------------------------------
📝 Want more Open Source content? Read our blog for the latest: www.tweag.io/blog
📈 We are big on composable software: functional, typed, and immutable. You might have seen some of our work in your own favourite technical community. Check out our Github page: github.com/tweag
💻 Tweag, a Modus Create company, is a software innovation lab with the mission to advance software craftsmanship and developer experience with meaningful open-source contributions. Our research helps fuel some of the most exciting projects at the world’s largest enterprises.
🌐 Tweag’s team of engineers are behind today’s boldest innovations in machine learning, distributed computing, and biotech. Visit www.tweag.io to learn more.
Негізгі бет Paolo Perrone - Markov categories, Bayesian inversions, and statistical experiments
Пікірлер