Typed Abstractions for Causal Probabilistic Programming
summary
A Haskell library reimplements Cairo’s causal features with concrete types, extracting a cosmonad (a reader monad transformer over a probability monad) to represent intervention maps and explicit intervention points as typed holes, and an applicative multial to eagerly twin worlds. The pipeline turns Pearl-style program transforms into environment manipulations, runs inference on the twinned model, and returns the counterfactual probability ~64% for the car example, demonstrating the abstractions compose and produce the expected numeric result.
AI-written summary · about
chapters (5)
- 0:00Typed causal programming intro
- 1:47Core abstractions and typed Cairo
- 6:20First-class causal Cairo
- 9:12Twinning with applicative multial
- 11:02Conclusion and future
related talks
24mHigher order programming with probabilistic effects: A model of stochastic memo...
38mProbabilistic programming and meta-programming in Clojure
43mA Practical Guide to Causality in Python (For The Perplexed)
46mBlueprints for a Universal Reasoning Machine
29mProbabilistic Programming: What It Is and How It Works
01h 03mA Personal Viewpoint on Probabilistic Programming
40mProbabilistic Programming for Augmented Intelligence
19mIncremental Computation for Efficient Programmable Inference in Probabilistic Programs