Deep neural networks are generally designed as a stack of differentiable layers, in which a prediction is obtained only after running the full stack. Recently, some contributions have proposed techniques to endow the networks with early exits, allowing to obtain predictions at intermediate points of the stack. These multi-output networks have a number of advantages, including (i) significant reductions of the inference time, (ii) reduced tendency to overfitting and vanishing gradients, and (iii) capability of being distributed over multi-tier computation platforms. In addition, they connect to the wider themes of biological plausibility and layered cognitive reasoning. In this paper, we provide a comprehensive introduction to this family of neural networks, by describing in a unified fashion the way these architectures can be designed, trained, and actually deployed in time-constrained scenarios. We also describe in-depth their application scenarios in 5G and Fog computing environments, as long as some of the open research questions connected to them.

Why should we add early exits to neural networks? / Scardapane, S.; Scarpiniti, M.; Baccarelli, E.; Uncini, A.. - In: COGNITIVE COMPUTATION. - ISSN 1866-9956. - 12:5(2020), pp. 954-966. [10.1007/s12559-020-09734-4]

Why should we add early exits to neural networks?

Scardapane S.
;
Scarpiniti M.;Baccarelli E.;Uncini A.
2020

Abstract

Deep neural networks are generally designed as a stack of differentiable layers, in which a prediction is obtained only after running the full stack. Recently, some contributions have proposed techniques to endow the networks with early exits, allowing to obtain predictions at intermediate points of the stack. These multi-output networks have a number of advantages, including (i) significant reductions of the inference time, (ii) reduced tendency to overfitting and vanishing gradients, and (iii) capability of being distributed over multi-tier computation platforms. In addition, they connect to the wider themes of biological plausibility and layered cognitive reasoning. In this paper, we provide a comprehensive introduction to this family of neural networks, by describing in a unified fashion the way these architectures can be designed, trained, and actually deployed in time-constrained scenarios. We also describe in-depth their application scenarios in 5G and Fog computing environments, as long as some of the open research questions connected to them.
2020
conditional computation; deep learning; distributed optimization; early exit; fog computing
01 Pubblicazione su rivista::01a Articolo in rivista
Why should we add early exits to neural networks? / Scardapane, S.; Scarpiniti, M.; Baccarelli, E.; Uncini, A.. - In: COGNITIVE COMPUTATION. - ISSN 1866-9956. - 12:5(2020), pp. 954-966. [10.1007/s12559-020-09734-4]
File allegati a questo prodotto
File Dimensione Formato  
Scardapane_Why-should_2020.pdf

solo gestori archivio

Tipologia: Versione editoriale (versione pubblicata con il layout dell'editore)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 697.52 kB
Formato Adobe PDF
697.52 kB Adobe PDF   Contatta l'autore
Scardapane_preprint_Why_2020.pdf

solo gestori archivio

Tipologia: Documento in Pre-print (manoscritto inviato all'editore, precedente alla peer review)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 636.21 kB
Formato Adobe PDF
636.21 kB Adobe PDF   Contatta l'autore

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11573/1438204
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 73
  • ???jsp.display-item.citation.isi??? 49
social impact