Towards Compiling Higher-Order Programs to Bayesian Networks
summary
Claims an effective compilation from a first‑order probabilistic lambda fragment to Bayesian networks, with a path to higher‑order and recursion, by running a parallel Krivine Abstract Machine that builds the Bayesian network while evaluating. Implementation uses an environment plus a global graph: tuples spawn threads, sample/case create named nodes (position‑based), edges track data flow; correctness is proved by showing semantic equivalence and machine confluence. Also outlines translation of resulting Bayesian networks to quantum circuits for a quadratic inference speedup and pruning to cut qubit needs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:23Bayesian networks background
- 2:56First-order to Bayesian networks
- 4:44Abstract machine for variables
- 6:30Abstract machine rules
- 8:42Correctness proof
- 9:23Motivation for Bayesian networks
- 10:49Quantum circuits connection
- 11:52Q&A
related talks
16mHigher Order Bayesian Networks, Exactly
18mLearning Abstraction Selection for Bayesian Program Analysis
24mExact Bayesian Inference for Loopy Probabilistic Programs Using Generating Functions
01h 03mAn Overview of Probabilistic Programming
01h 03mA Personal Viewpoint on Probabilistic Programming
38mProbabilistic programming and meta-programming in Clojure
24mHigher order programming with probabilistic effects: A model of stochastic memo...
26mAffine Monads and Lazy Structures for Bayesian Programming