A Welcome to Causal Probabilistic Programming
summary
Probabilistic programming is extended to answer interventional and counterfactual queries using concrete tools: Cairo on Pyro (with a Haskell port) for effect-handler implementations and a prototype graded monad design that exposes intervention points via an intervention primitive int. Grading solves many static rung‑two cases; dynamic approaches use traces, Omega‑style dynamic scoping or effect handlers. A formal proof of the backdoor adjustment runs to about 200 lines in the Unimath/Rock proof assistant; simple examples use 10%/90% error rates to show differing counterfactual behaviour.
AI-written summary · about
chapters (9)
- 0:00Causal probabilistic programming
- 1:28What is causality and why it matters
- 9:53Pearl’s three rungs of causality
- 23:34Defining causal probabilistic programming
- 28:09Goals and challenges in causal PP
- 32:42Approaches to interventions
- 44:07Theoretical foundations and category theory
- 51:15Formal proofs in Unimath
- 54:06Closing and future directions
related talks
31mProbabilistic Programming: The What, Why and How
01h 03mA Personal Viewpoint on Probabilistic Programming
38mProbabilistic programming and meta-programming in Clojure
47mThe State of the Art for Probabilistic Programming
01h 03mAn Overview of Probabilistic Programming
24mHigher order programming with probabilistic effects: A model of stochastic memo...
43mA Practical Guide to Causality in Python (For The Perplexed)
40mProbabilistic Programming for Augmented Intelligence