Deep learning with differential Gaussian process flows

Pashupati Hegde; Markus Heinonen; Harri Lahdesmaki; Samuel Kaski

Deep learning with differential Gaussian process flows

Pashupati Hegde
, Markus Heinonen
, Harri Lahdesmaki
, Samuel Kaski

Computer Science Teaching

Helsinki Institute for Information Technology

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

Abstract

We propose a novel deep learning paradigm of differential flows that learn a stochastic differential equation transformations of inputs prior to a standard classification or regression function. The key property of differential Gaussian processes is the warping of inputs through infinitely deep, but infinitesimal, differential fields, that generalise discrete layers into a dynamical system. We demonstrate excellent results as compared to deep Gaussian processes and Bayesian neural networks.

Original language	English
Title of host publication	Proceedings of Machine Learning Research: Artificial Intelligence and Statistics 2019
Pages	1812-1821
Volume	89
Publication status	Published - 16 Apr 2019
Event	22nd International Conference on Artificial Intelligence and Statistics - Naha, Japan Duration: 16 Apr 2019 → 18 Apr 2019

Publication series

Name	Proceedings of Machine Learning Research
Publisher	MLResearchPress
ISSN (Electronic)	2640-3498

Conference

Conference	22nd International Conference on Artificial Intelligence and Statistics
Abbreviated title	AISTATS 2019
Country/Territory	Japan
City	Naha
Period	16/04/19 → 18/04/19

Keywords

Gaussian processes

Access to Document

http://proceedings.mlr.press/v89/hegde19a.html

Cite this

@inproceedings{6ad9bf098d364805bcaf419e8ebd8f90,

title = "Deep learning with differential Gaussian process flows",

abstract = "We propose a novel deep learning paradigm of differential flows that learn a stochastic differential equation transformations of inputs prior to a standard classification or regression function. The key property of differential Gaussian processes is the warping of inputs through infinitely deep, but infinitesimal, differential fields, that generalise discrete layers into a dynamical system. We demonstrate excellent results as compared to deep Gaussian processes and Bayesian neural networks.",

keywords = "Gaussian processes",

author = "Pashupati Hegde and Markus Heinonen and Harri Lahdesmaki and Samuel Kaski",

year = "2019",

month = apr,

day = "16",

language = "English",

volume = "89",

series = "Proceedings of Machine Learning Research",

publisher = "MLResearchPress",

pages = "1812--1821",

booktitle = "Proceedings of Machine Learning Research: Artificial Intelligence and Statistics 2019",

note = "22nd International Conference on<br/>Artificial Intelligence and Statistics, AISTATS 2019 ; Conference date: 16-04-2019 Through 18-04-2019",

}

Hegde, P, Heinonen, M, Lahdesmaki, H & Kaski, S 2019, Deep learning with differential Gaussian process flows. in Proceedings of Machine Learning Research: Artificial Intelligence and Statistics 2019. vol. 89, Proceedings of Machine Learning Research, pp. 1812-1821, 22nd International Conference on
Artificial Intelligence and Statistics, Naha, Japan, 16/04/19. <http://proceedings.mlr.press/v89/hegde19a.html>

Deep learning with differential Gaussian process flows. / Hegde, Pashupati; Heinonen, Markus; Lahdesmaki, Harri et al.
Proceedings of Machine Learning Research: Artificial Intelligence and Statistics 2019. Vol. 89 2019. p. 1812-1821 (Proceedings of Machine Learning Research).

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Deep learning with differential Gaussian process flows

AU - Hegde, Pashupati

AU - Heinonen, Markus

AU - Lahdesmaki, Harri

AU - Kaski, Samuel

PY - 2019/4/16

Y1 - 2019/4/16

N2 - We propose a novel deep learning paradigm of differential flows that learn a stochastic differential equation transformations of inputs prior to a standard classification or regression function. The key property of differential Gaussian processes is the warping of inputs through infinitely deep, but infinitesimal, differential fields, that generalise discrete layers into a dynamical system. We demonstrate excellent results as compared to deep Gaussian processes and Bayesian neural networks.

AB - We propose a novel deep learning paradigm of differential flows that learn a stochastic differential equation transformations of inputs prior to a standard classification or regression function. The key property of differential Gaussian processes is the warping of inputs through infinitely deep, but infinitesimal, differential fields, that generalise discrete layers into a dynamical system. We demonstrate excellent results as compared to deep Gaussian processes and Bayesian neural networks.

KW - Gaussian processes

M3 - Conference contribution

VL - 89

T3 - Proceedings of Machine Learning Research

SP - 1812

EP - 1821

BT - Proceedings of Machine Learning Research: Artificial Intelligence and Statistics 2019

T2 - 22nd International Conference on<br/>Artificial Intelligence and Statistics

Y2 - 16 April 2019 through 18 April 2019

ER -

Deep learning with differential Gaussian process flows

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Cite this