Automatic Differentiation Variational Inference

Alp Kucukelbir, David M. Blei, Andrew Gelman, Rajesh Ranganath, Dustin Tran

Research output: Contribution to journalArticle

Abstract

Probabilistic modeling is iterative. A scientist posits a simple model, fits it to her data, refines it according to her analysis, and repeats. However, fitting complex models to large data is a bottleneck in this process. Deriving algorithms for new models can be both mathematically and computationally challenging, which makes it difficult to efficiently cycle through the steps. To this end, we develop automatic differentiation variational inference (advi). Using our method, the scientist only provides a probabilistic model and a dataset, nothing else. advi automatically derives an efficient variational inference algorithm, freeing the scientist to refine and explore many models. advi supports a broad class of models-no conjugacy assumptions are required. We study advi across ten modern probabilistic models and apply it to a dataset with millions of observations. We deploy advi as part of Stan, a probabilistic programming system.

Original languageEnglish (US)
Pages (from-to)1-45
Number of pages45
JournalJournal of Machine Learning Research
Volume18
StatePublished - Jan 1 2017

Fingerprint

Automatic Differentiation
Probabilistic Model
Computer systems programming
Probabilistic Programming
Probabilistic Modeling
Model
Conjugacy
Large Data
Cycle

Keywords

  • Approximate inference
  • Bayesian inference
  • Probabilistic programming

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Software
  • Statistics and Probability
  • Artificial Intelligence

Cite this

Kucukelbir, A., Blei, D. M., Gelman, A., Ranganath, R., & Tran, D. (2017). Automatic Differentiation Variational Inference. Journal of Machine Learning Research, 18, 1-45.

Automatic Differentiation Variational Inference. / Kucukelbir, Alp; Blei, David M.; Gelman, Andrew; Ranganath, Rajesh; Tran, Dustin.

In: Journal of Machine Learning Research, Vol. 18, 01.01.2017, p. 1-45.

Research output: Contribution to journalArticle

Kucukelbir, A, Blei, DM, Gelman, A, Ranganath, R & Tran, D 2017, 'Automatic Differentiation Variational Inference', Journal of Machine Learning Research, vol. 18, pp. 1-45.
Kucukelbir A, Blei DM, Gelman A, Ranganath R, Tran D. Automatic Differentiation Variational Inference. Journal of Machine Learning Research. 2017 Jan 1;18:1-45.
Kucukelbir, Alp ; Blei, David M. ; Gelman, Andrew ; Ranganath, Rajesh ; Tran, Dustin. / Automatic Differentiation Variational Inference. In: Journal of Machine Learning Research. 2017 ; Vol. 18. pp. 1-45.
@article{befbf1427e7c480c964a20896a41b9a7,
title = "Automatic Differentiation Variational Inference",
abstract = "Probabilistic modeling is iterative. A scientist posits a simple model, fits it to her data, refines it according to her analysis, and repeats. However, fitting complex models to large data is a bottleneck in this process. Deriving algorithms for new models can be both mathematically and computationally challenging, which makes it difficult to efficiently cycle through the steps. To this end, we develop automatic differentiation variational inference (advi). Using our method, the scientist only provides a probabilistic model and a dataset, nothing else. advi automatically derives an efficient variational inference algorithm, freeing the scientist to refine and explore many models. advi supports a broad class of models-no conjugacy assumptions are required. We study advi across ten modern probabilistic models and apply it to a dataset with millions of observations. We deploy advi as part of Stan, a probabilistic programming system.",
keywords = "Approximate inference, Bayesian inference, Probabilistic programming",
author = "Alp Kucukelbir and Blei, {David M.} and Andrew Gelman and Rajesh Ranganath and Dustin Tran",
year = "2017",
month = "1",
day = "1",
language = "English (US)",
volume = "18",
pages = "1--45",
journal = "Journal of Machine Learning Research",
issn = "1532-4435",
publisher = "Microtome Publishing",

}

TY - JOUR

T1 - Automatic Differentiation Variational Inference

AU - Kucukelbir, Alp

AU - Blei, David M.

AU - Gelman, Andrew

AU - Ranganath, Rajesh

AU - Tran, Dustin

PY - 2017/1/1

Y1 - 2017/1/1

N2 - Probabilistic modeling is iterative. A scientist posits a simple model, fits it to her data, refines it according to her analysis, and repeats. However, fitting complex models to large data is a bottleneck in this process. Deriving algorithms for new models can be both mathematically and computationally challenging, which makes it difficult to efficiently cycle through the steps. To this end, we develop automatic differentiation variational inference (advi). Using our method, the scientist only provides a probabilistic model and a dataset, nothing else. advi automatically derives an efficient variational inference algorithm, freeing the scientist to refine and explore many models. advi supports a broad class of models-no conjugacy assumptions are required. We study advi across ten modern probabilistic models and apply it to a dataset with millions of observations. We deploy advi as part of Stan, a probabilistic programming system.

AB - Probabilistic modeling is iterative. A scientist posits a simple model, fits it to her data, refines it according to her analysis, and repeats. However, fitting complex models to large data is a bottleneck in this process. Deriving algorithms for new models can be both mathematically and computationally challenging, which makes it difficult to efficiently cycle through the steps. To this end, we develop automatic differentiation variational inference (advi). Using our method, the scientist only provides a probabilistic model and a dataset, nothing else. advi automatically derives an efficient variational inference algorithm, freeing the scientist to refine and explore many models. advi supports a broad class of models-no conjugacy assumptions are required. We study advi across ten modern probabilistic models and apply it to a dataset with millions of observations. We deploy advi as part of Stan, a probabilistic programming system.

KW - Approximate inference

KW - Bayesian inference

KW - Probabilistic programming

UR - http://www.scopus.com/inward/record.url?scp=85016397096&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85016397096&partnerID=8YFLogxK

M3 - Article

VL - 18

SP - 1

EP - 45

JO - Journal of Machine Learning Research

JF - Journal of Machine Learning Research

SN - 1532-4435

ER -