Martin Jankowiak - Brief Introduction to Probabilistic Programming

Recorded at the ML in PL 2019 Conference, the University of Warsaw, 22-24 November 2019.
Martin Jankowiak (Uber AI Labs)
Slides available at docs.mlinpl.org/conference/201...
Abstract:
Probabilistic models offer a compelling methodology for reasoning about an uncertain world. Programming languages are powerful tools for specifying deterministic computations. Their synthesis--probabilistic programming languages (PPLs)--promises a unified and (partially) automated approach to specifying and reasoning about complex models. We give an introduction to PPLs, with examples drawn from economics and natural science serving as motivation. For concreteness we illustrate all our examples using the Pyro PPL.
Relevant links:
pyro.ai/
eng.uber.com/oed-pyro-release/
An Introduction to Probabilistic Programming arxiv.org/abs/1809.10756

[@sinan_islam]

Probabilistic Programming is same as Bayesian Statistics?

[@sinan_islam]

@Hobbesian Thinker Thank you!

[@couldntfindafreename]

Naming conflict: pyro was Python Remote Objects 10 years ago...

[@kevincannon2269]

Naming conflict #2: H0 is also used to represent the null hypothesis

[@jeffreyalidochair]

why, at 19:15, is the visualization of the slop curved? the model seems to be a degree 1 polynomial

[@jeffreyalidochair]

oh its log(ax+b) i guess

[@peabrane8067]

12:05 😂😂😂😂

[@MusixPro4u]

35:21 the formula for the posterior is wrong, isn't it? It should be p(theta|y,d) instead of p(y,theta|d).

[@matthewlvk7366]

I think that's just simpe formulation of conditional probability, the starting formular of Bayes' Theorem
first equation in en.wikipedia.org/wiki/Conditional_probability