Using Differentiable Programming for Flexible Statistical Modeling.

Differentiable programming has recently received much interest as a paradigm that facilitates taking gradients of computer programs. While the corresponding flexible gradient-based optimization approaches so far have been used predominantly for deep learning or enriching the latter with modeling com...

Descripción completa

Detalles Bibliográficos
Publicado en:American Statistician Vol. 76; no. 3; pp. 270 - 280
Autores principales: Hackenberg, Maren, Grodd, Marlon, Kreutz, Clemens, Fischer, Martina, Esins, Janina, Grabenhenrich, Linus, Karagiannidis, Christian, Binder, Harald
Formato: Artículo
Publicado: Taylor & Francis Ltd Aug2022
Materias:
Acceso en línea:Ver este registro en EBSCOhost
fields @attributes:
  recordID: 1
pdfLink:
plink: https://search.ebscohost.com/login.aspx?direct=true&db=ssf&AN=158065777&site=ehost-live
header:
  @attributes:
    shortDbName: ssf
    uiTerm: 158065777
    longDbName: Social Sciences Full Text (H.W. Wilson)
    uiTag: AN
  controlInfo:
    bkinfo:
    jinfo:
      jid:
        00031305
        STT
      jtl: American Statistician
      issn: 00031305
      maglogo: Y
    pubinfo:
      dt: Aug2022
      vid: 76
      iid: 3
      pid: 377
      pub: Taylor & Francis Ltd
    artinfo:
      ui:
        158065777
        10.1080/00031305.2021.2002189
      ppf: 270
      ppct: 10
      formats:
      tig:
        atl: Using Differentiable Programming for Flexible Statistical Modeling.
      aug:
        au:
          Hackenberg, Maren
          Grodd, Marlon
          Kreutz, Clemens
          Fischer, Martina
          Esins, Janina
          Grabenhenrich, Linus
          Karagiannidis, Christian
          Binder, Harald
        affil:
          Institute of Medical Biometry and Statistics, Faculty of Medicine and Medical Center, University of Freiburg, Germany
          Robert-Koch-Institut, Berlin, Germany
          Department of Pneumology and Critical Care Medicine, Cologne-Merheim Hospital, ARDS and ECMO Center, Kliniken der Stadt Köln, Witten/Herdecke University Hospital, Cologne, Germany
      su:
        Time pressure
        Statistical models
        Deep learning
        Delay differential equations
        Automatic differentiation
        Computer software
      sug:
        subj:
          Time pressure
          Software publishers (except video game publishers)
          Computer and Computer Peripheral Equipment and Software Merchant Wholesalers
          Computer, computer peripheral and pre-packaged software merchant wholesalers
          Computer and software stores
          Statistical models
          Deep learning
          Delay differential equations
          Automatic differentiation
          Computer software
      keyword:
        Differential equations
        Machine learning
        Optimization
        Workflow
        Differential equations
        Machine learning
        Optimization
        Workflow
      ab: Differentiable programming has recently received much interest as a paradigm that facilitates taking gradients of computer programs. While the corresponding flexible gradient-based optimization approaches so far have been used predominantly for deep learning or enriching the latter with modeling components, we want to demonstrate that they can also be useful for statistical modeling per se, for example, for quick prototyping when classical maximum likelihood approaches are challenging or not feasible. In an application from a COVID-19 setting, we use differentiable programming to quickly build and optimize a flexible prediction model adapted to the data quality challenges at hand. Specifically, we develop a regression model, inspired by delay differential equations, that can bridge temporal gaps of observations in the central German registry of COVID-19 intensive care cases for predicting future demand. With this exemplary modeling challenge, we illustrate how differentiable programming can enable simple gradient-based optimization of the model by automatic differentiation. This allowed us to quickly prototype a model under time pressure that outperforms simpler benchmark models. We thus exemplify the potential of differentiable programming also outside deep learning applications to provide more options for flexible applied statistical modeling.
      pubtype: Academic Journal
      doctype: Article
      src: R
    language: English
    refInfo:
    copyright:
      @attributes:
        flag: N
    holdings:
      @attributes:
        islocal: N