Multitask prediction of organ dysfunction in the intensive care unit using sequential subnetwork routing.

Journal: Journal of the American Medical Informatics Association : JAMIA
PMID: 34151965

Abstract

OBJECTIVE: Multitask learning (MTL) using electronic health records allows concurrent prediction of multiple endpoints. MTL has shown promise in improving model performance and training efficiency; however, it often suffers from negative transfer - impaired learning if tasks are not appropriately selected. We introduce a sequential subnetwork routing (SeqSNR) architecture that uses soft parameter sharing to find related tasks and encourage cross-learning between them.

Authors

  • Subhrajit Roy
    IBM Research - Australia, 204 Lygon Street, 3053 Carlton, VIC, Australia.
  • Diana Mincu
    Google Health, London, United Kingdom.
  • Eric Loreaux
    Google Health, London, United Kingdom.
  • Anne Mottram
    DeepMind, London, UK.
  • Ivan Protsyuk
    Google Health, London, UK.
  • Natalie Harris
    Google Health, London, UK.
  • Yuan Xue
    Department of Nutrition and food hygiene, College of Public Health of Zhengzhou University, Zhengzhou, China, 450001. Electronic address: 962634470@qq.com.
  • Jessica Schrouff
    Laboratory of Behavioral & Cognitive Neuroscience, Stanford Human Intracranial Cognitive Electrophysiology Program (SHICEP), Department of Neurology & Neurological Sciences, Stanford University, Stanford, CA, USA; Department of Computer Science, University College London, United Kingdom.
  • Hugh Montgomery
    Institute of Sport, Exercise and Health, London, W1T 7HA, UK.
  • Alistair Connell
    Google Health, London, UK.
  • Nenad Tomasev
    DeepMind, London, EC4A 3TW, UK.
  • Alan Karthikesalingam
    Department of Outcomes Research, St George's Vascular Institute, London, SW17 0QT, United Kingdom.
  • Martin Seneviratne
    Stanford Center for Biomedical Informatics Research, Stanford, California 94305, USA.

Keywords

External Resources

Popular Topics
Recent Journals