Adaptive neural network architectures for power aware inference

Skyler Anderson, Nagadastagiri Challapalle, John Sampson, Vijaykrishnan Narayanan

Research output: Contribution to journalArticle

Abstract

When dealing with edge devices, diverse power and compute constraints impose tradeoffs among performance, accuracy, and energy requirements during inference. Thus, ensemble learning (multiple learning algorithms to obtain better predictive performance than could be obtained from any of the constituent learning algorithms alone) has been a popular emergent paradigm for accuracy improvement in such scenarios. In cases where powering of the edge devices varies, various ensemble models can be dynamically adopted to obtain the best performance-to-power ratio. This article describes how an agnostic to the base network method can adaptively boost the performance depending on energy availability. - Theocharis Theocharides, University of Cyprus - Muhammad Shafique, Technische Universitat Wien.

Original languageEnglish (US)
Article number8868173
Pages (from-to)66-75
Number of pages10
JournalIEEE Design and Test
Volume37
Issue number2
DOIs
StatePublished - Apr 2020

All Science Journal Classification (ASJC) codes

  • Software
  • Hardware and Architecture
  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'Adaptive neural network architectures for power aware inference'. Together they form a unique fingerprint.

  • Cite this