Off-the-shelf deep learning is not enough, and requires parsimony, Bayesianity, and causality

Rama K. Vasudevan; Maxim Ziatdinov; Lukas Vlcek; Sergei V. Kalinin

doi:10.1038/s41524-020-00487-0

Off-the-shelf deep learning is not enough, and requires parsimony, Bayesianity, and causality

Rama K. Vasudevan, Maxim Ziatdinov, Lukas Vlcek, Sergei V. Kalinin

Research output: Contribution to journal › Review article › peer-review

38 Scopus citations

Abstract

Deep neural networks (‘deep learning’) have emerged as a technology of choice to tackle problems in speech recognition, computer vision, finance, etc. However, adoption of deep learning in physical domains brings substantial challenges stemming from the correlative nature of deep learning methods compared to the causal, hypothesis driven nature of modern science. We argue that the broad adoption of Bayesian methods incorporating prior knowledge, development of solutions with incorporated physical constraints and parsimonious structural descriptors and generative models, and ultimately adoption of causal models, offers a path forward for fundamental and applied research.

Original language	English
Article number	16
Journal	npj Computational Materials
Volume	7
Issue number	1
DOIs	https://doi.org/10.1038/s41524-020-00487-0
State	Published - Dec 2021
Externally published	Yes

Funding

The work was supported by the U.S. Department of Energy, Office of Science, Materials Sciences and Engineering Division (S.V.K., L.V., R. K. V.). Research was conducted at the Center for Nanophase Materials Sciences, which also provided support (M.Z.) and is a US DOE Office of Science User Facility.

Access to Document

10.1038/s41524-020-00487-0

Cite this

@article{a781eea03af24746b762f78bf856eb46,

title = "Off-the-shelf deep learning is not enough, and requires parsimony, Bayesianity, and causality",

abstract = "Deep neural networks ({\textquoteleft}deep learning{\textquoteright}) have emerged as a technology of choice to tackle problems in speech recognition, computer vision, finance, etc. However, adoption of deep learning in physical domains brings substantial challenges stemming from the correlative nature of deep learning methods compared to the causal, hypothesis driven nature of modern science. We argue that the broad adoption of Bayesian methods incorporating prior knowledge, development of solutions with incorporated physical constraints and parsimonious structural descriptors and generative models, and ultimately adoption of causal models, offers a path forward for fundamental and applied research.",

author = "Vasudevan, {Rama K.} and Maxim Ziatdinov and Lukas Vlcek and Kalinin, {Sergei V.}",

note = "Publisher Copyright: {\textcopyright} 2021, This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply.",

year = "2021",

month = dec,

doi = "10.1038/s41524-020-00487-0",

language = "English",

volume = "7",

journal = "npj Computational Materials",

issn = "2057-3960",

publisher = "Nature Research",

number = "1",

}

TY - JOUR

T1 - Off-the-shelf deep learning is not enough, and requires parsimony, Bayesianity, and causality

AU - Vasudevan, Rama K.

AU - Ziatdinov, Maxim

AU - Vlcek, Lukas

AU - Kalinin, Sergei V.

PY - 2021/12

Y1 - 2021/12

N2 - Deep neural networks (‘deep learning’) have emerged as a technology of choice to tackle problems in speech recognition, computer vision, finance, etc. However, adoption of deep learning in physical domains brings substantial challenges stemming from the correlative nature of deep learning methods compared to the causal, hypothesis driven nature of modern science. We argue that the broad adoption of Bayesian methods incorporating prior knowledge, development of solutions with incorporated physical constraints and parsimonious structural descriptors and generative models, and ultimately adoption of causal models, offers a path forward for fundamental and applied research.

AB - Deep neural networks (‘deep learning’) have emerged as a technology of choice to tackle problems in speech recognition, computer vision, finance, etc. However, adoption of deep learning in physical domains brings substantial challenges stemming from the correlative nature of deep learning methods compared to the causal, hypothesis driven nature of modern science. We argue that the broad adoption of Bayesian methods incorporating prior knowledge, development of solutions with incorporated physical constraints and parsimonious structural descriptors and generative models, and ultimately adoption of causal models, offers a path forward for fundamental and applied research.

UR - http://www.scopus.com/inward/record.url?scp=85099876938&partnerID=8YFLogxK

U2 - 10.1038/s41524-020-00487-0

DO - 10.1038/s41524-020-00487-0

M3 - Review article

AN - SCOPUS:85099876938

SN - 2057-3960

VL - 7

JO - npj Computational Materials

JF - npj Computational Materials

IS - 1

M1 - 16

ER -

Off-the-shelf deep learning is not enough, and requires parsimony, Bayesianity, and causality

Abstract

Funding

Access to Document

Other files and links

Fingerprint

Cite this