HackPPL: a universal probabilistic programming language

Ai, Jessica; Arora, Neha; Gokkaya, Beliz U.; Jiang, Thomas; Kubendran, Anitha; Kumar, Arun; Tingley, Michael; Torabi, Narjes

doi:10.1145/3315508.3329974

Cited by 6 publications

(7 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Type operators prescribe guidance protocols for procedures by parameterizing with a continuation type that describes the guidance protocol after a procedure call. A procedure signature 1 2 | (a : ); (b : ) types a procedure that takes a parameter of type 1 , returns a result of type 2 , consumes a channel , and provides a channel , such that if the guidance protocols for and after a procedure call are and , respectively, then the guidance protocols for and before the procedure call are [ ] and [ ], respectively. A type definition typedef( .…”

Section: Guide Typesmentioning

confidence: 99%

“…which is obtained by factoring out the observation @𝑧 from the support of the unconditioned model shown in (1). Fig.…”

Section: Bayesian Inferencementioning

confidence: 99%

“…A guidance trace 𝜎 is a finite sequence of guidance messages exchanged on a channel; each guidance message has the form val P (𝑣) (resp., dir P (𝑣)) for a sample value 𝑣 (resp., a branch selection 𝑣) from the provider to the consumer, the form val C (𝑣) (resp., dir C (𝑣)) for a sample value 𝑣 (resp., a branch selection 𝑣) from the consumer to the provider, or a procedure-call indicator fold. 1 The evaluation judgments for commands have the form 𝑉 | (a : 𝜎 𝑎 ); (b : 𝜎 𝑏 ) ⊢ 𝑚 ⇓ 𝑤 𝑣, where 𝑉 is an environment, 𝑚 is a command that consumes channel 𝑎 and provides channel 𝑏, 𝜎 𝑎 and 𝜎 𝑏 are guidance traces on the channels, 𝑣 is the evaluation result, and 𝑤 ≥ 0 is a weight that expresses how likely the guidance traces are. Intuitively, a probabilistic program specifies a probability distribution on guidance traces, and the weights represent probability densities with respect to the distribution.…”

Section: A Coroutine-based Pplmentioning

confidence: 99%

“…Probabilistic programming languages (PPLs) [1,13,23,25,26,49,52,57,62] provide a flexible way of describing statistical models and automatically performing Bayesian inference: a method for inferring the posterior of a statistical model from observed data. Bayesian inference accounts for uncertainty in latent variables that produce the observed data.…”

Section: Introductionmentioning

confidence: 99%

“…( =⇒ ) Assume P m ( ℓ , ) ≠ 0. Thus, there exists 1 such that ∅ | (latent : ℓ ); (obs : ) ⊢ m ⇓ 1 1 , where 1 def = P m ( ℓ , ). Because is -free, we can apply Thm.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Sound probabilistic inference via guide types

Wang

Hoffmann

Reps³

2021

Proceedings of the 42nd ACM SIGPLAN International Conference on Programming Language Design and Implementation

View full text Add to dashboard Cite

Probabilistic programming languages aim to describe and automate Bayesian modeling and inference. Modern languages support programmable inference, which allows users to customize inference algorithms by incorporating guide programs to improve inference performance. For Bayesian inference to be sound, guide programs must be compatible with model programs. One pervasive but challenging condition for model-guide compatibility is absolute continuity, which requires that the model and guide programs define probability distributions with the same support.This paper presents a new probabilistic programming language that guarantees absolute continuity, and features general programming constructs, such as branching and recursion. Model and guide programs are implemented as coroutines that communicate with each other to synchronize the set of random variables they sample during their execution. Novel guide types describe and enforce communication protocols between coroutines. If the model and guide are well-typed using the same protocol, then they are guaranteed to enjoy absolute continuity. An efficient algorithm infers guide types from code so that users do not have to specify the types. The new programming language is evaluated with an implementation that includes the type-inference algorithm and a prototype compiler that targets Pyro. Experiments show that our language is capable of expressing a variety of probabilistic models with nontrivial control flow and recursion, and that the coroutine-based computation does not introduce significant overhead in actual Bayesian inference.

show abstract

Section: Guide Typesmentioning

confidence: 99%

“…which is obtained by factoring out the observation @𝑧 from the support of the unconditioned model shown in (1). Fig.…”

Section: Bayesian Inferencementioning

confidence: 99%

Section: A Coroutine-based Pplmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…( =⇒ ) Assume P m ( ℓ , ) ≠ 0. Thus, there exists 1 such that ∅ | (latent : ℓ ); (obs : ) ⊢ m ⇓ 1 1 , where 1 def = P m ( ℓ , ). Because is -free, we can apply Thm.…”

mentioning

confidence: 99%

See 3 more Smart Citations