Compiling discrete probabilistic programs for vectorized exact inference

Jingwen Pan; Amir Shaikhha

doi:10.1145/3578360.3580258

Compiling discrete probabilistic programs for vectorized exact inference

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract / Description of output

Probabilistic programming languages (PPLs) are essential for reasoning under uncertainty. Even though many real-world probabilistic programs involve discrete distributions, the state-of-the-art PPLs are suboptimal for a large class of tasks dealing with such distributions. In this paper, we propose BayesTensor, a tensor-based probabilistic programming framework. By generating tensor algebra code from probabilistic programs, BayesTensor takes advantage of the highly-tuned vectorized implementations of tensor processing frameworks. Our experiments show that BayesTensor outperforms the state-of-the-art frameworks in a variety of discrete probabilistic programs, inference over Bayesian Networks, and real-world probabilistic programs employed in data processing systems.

Original language	English
Title of host publication	CC 2023 - Proceedings of the 32nd ACM SIGPLAN International Conference on Compiler Construction
Editors	Clark Verbrugge, Ondrej Lhotak, Xipeng Shen
Publisher	ACM
Pages	13-24
Number of pages	12
ISBN (Electronic)	9798400700880
DOIs	https://doi.org/10.1145/3578360.3580258
Publication status	Published - 17 Feb 2023
Event	32nd ACM SIGPLAN International Conference on Compiler Construction - Montreal, Canada Duration: 25 Feb 2023 → 26 Feb 2023

Conference

Conference	32nd ACM SIGPLAN International Conference on Compiler Construction
Abbreviated title	CC 2023
Country/Territory	Canada
City	Montreal
Period	25/02/23 → 26/02/23

Keywords / Materials (for Non-textual outputs)

cardinality estimation
discrete distribution
probabilistic programming
tensor algebra

Access to Document

10.1145/3578360.3580258

Cite this

@inproceedings{0df71045dcf246269ced980c314edca6,

title = "Compiling discrete probabilistic programs for vectorized exact inference",

abstract = "Probabilistic programming languages (PPLs) are essential for reasoning under uncertainty. Even though many real-world probabilistic programs involve discrete distributions, the state-of-the-art PPLs are suboptimal for a large class of tasks dealing with such distributions. In this paper, we propose BayesTensor, a tensor-based probabilistic programming framework. By generating tensor algebra code from probabilistic programs, BayesTensor takes advantage of the highly-tuned vectorized implementations of tensor processing frameworks. Our experiments show that BayesTensor outperforms the state-of-the-art frameworks in a variety of discrete probabilistic programs, inference over Bayesian Networks, and real-world probabilistic programs employed in data processing systems.",

keywords = "cardinality estimation, discrete distribution, probabilistic programming, tensor algebra",

author = "Jingwen Pan and Amir Shaikhha",

year = "2023",

month = feb,

day = "17",

doi = "10.1145/3578360.3580258",

language = "English",

pages = "13--24",

editor = "Clark Verbrugge and Ondrej Lhotak and Xipeng Shen",

booktitle = "CC 2023 - Proceedings of the 32nd ACM SIGPLAN International Conference on Compiler Construction",

publisher = "ACM",

note = "32nd ACM SIGPLAN International Conference on Compiler Construction, CC 2023 ; Conference date: 25-02-2023 Through 26-02-2023",

}

Pan, J & Shaikhha, A 2023, Compiling discrete probabilistic programs for vectorized exact inference. in C Verbrugge, O Lhotak & X Shen (eds), CC 2023 - Proceedings of the 32nd ACM SIGPLAN International Conference on Compiler Construction. ACM, pp. 13-24, 32nd ACM SIGPLAN International Conference on Compiler Construction, Montreal, Quebec, Canada, 25/02/23. https://doi.org/10.1145/3578360.3580258

TY - GEN

T1 - Compiling discrete probabilistic programs for vectorized exact inference

AU - Pan, Jingwen

AU - Shaikhha, Amir

PY - 2023/2/17

Y1 - 2023/2/17

N2 - Probabilistic programming languages (PPLs) are essential for reasoning under uncertainty. Even though many real-world probabilistic programs involve discrete distributions, the state-of-the-art PPLs are suboptimal for a large class of tasks dealing with such distributions. In this paper, we propose BayesTensor, a tensor-based probabilistic programming framework. By generating tensor algebra code from probabilistic programs, BayesTensor takes advantage of the highly-tuned vectorized implementations of tensor processing frameworks. Our experiments show that BayesTensor outperforms the state-of-the-art frameworks in a variety of discrete probabilistic programs, inference over Bayesian Networks, and real-world probabilistic programs employed in data processing systems.

AB - Probabilistic programming languages (PPLs) are essential for reasoning under uncertainty. Even though many real-world probabilistic programs involve discrete distributions, the state-of-the-art PPLs are suboptimal for a large class of tasks dealing with such distributions. In this paper, we propose BayesTensor, a tensor-based probabilistic programming framework. By generating tensor algebra code from probabilistic programs, BayesTensor takes advantage of the highly-tuned vectorized implementations of tensor processing frameworks. Our experiments show that BayesTensor outperforms the state-of-the-art frameworks in a variety of discrete probabilistic programs, inference over Bayesian Networks, and real-world probabilistic programs employed in data processing systems.

KW - cardinality estimation

KW - discrete distribution

KW - probabilistic programming

KW - tensor algebra

UR - http://www.scopus.com/inward/record.url?scp=85149187062&partnerID=8YFLogxK

U2 - 10.1145/3578360.3580258

DO - 10.1145/3578360.3580258

M3 - Conference contribution

AN - SCOPUS:85149187062

SP - 13

EP - 24

BT - CC 2023 - Proceedings of the 32nd ACM SIGPLAN International Conference on Compiler Construction

A2 - Verbrugge, Clark

A2 - Lhotak, Ondrej

A2 - Shen, Xipeng

PB - ACM

T2 - 32nd ACM SIGPLAN International Conference on Compiler Construction

Y2 - 25 February 2023 through 26 February 2023

ER -

Compiling discrete probabilistic programs for vectorized exact inference

Abstract / Description of output

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Other files and links

Fingerprint

Cite this