Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

Differentiable Branching in Deep Networks for Fast Inference

Abstract: 

In this paper, we consider the design of deep neural networks augmented with multiple auxiliary classifiers departing from the main (backbone) network. These classifiers can be used to perform early-exit from the network at various layers, making them convenient for energy-constrained applications such as IoT, embedded devices, or Fog computing. However, designing an optimized early-exit strategy is a difficult task, generally requiring a large amount of manual fine-tuning. In this paper, we propose a way to jointly optimize this strategy together with the branches, providing an end-to-end trainable algorithm for this emerging class of neural networks. We achieve this by replacing the original output of the branches with a ‘soft’, differentiable approximation. In addition, we also propose a regularization approach to trade-off the computational efficiency of the early-exit strategy with respect to the overall classification accuracy. We evaluate our proposed design approach on a set of image classification benchmarks, showing significant gains in accuracy and inference time.

up
0 users have voted:

Paper Details

Authors:
Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini
Submitted On:
5 June 2020 - 4:28am
Short Link:
Type:
Presentation Slides
Event:
Presenter's Name:
Danilo Comminiello
Paper Code:
1402
Document Year:
2020
Cite

Document Files

Paper #1402_Presentation.pdf

(18)

Subscribe

[1] Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini, "Differentiable Branching in Deep Networks for Fast Inference", IEEE SigPort, 2020. [Online]. Available: http://sigport.org/5453. Accessed: Jul. 09, 2020.
@article{5453-20,
url = {http://sigport.org/5453},
author = {Simone Scardapane; Danilo Comminiello; Michele Scarpiniti; Enzo Baccarelli; Aurelio Uncini },
publisher = {IEEE SigPort},
title = {Differentiable Branching in Deep Networks for Fast Inference},
year = {2020} }
TY - EJOUR
T1 - Differentiable Branching in Deep Networks for Fast Inference
AU - Simone Scardapane; Danilo Comminiello; Michele Scarpiniti; Enzo Baccarelli; Aurelio Uncini
PY - 2020
PB - IEEE SigPort
UR - http://sigport.org/5453
ER -
Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini. (2020). Differentiable Branching in Deep Networks for Fast Inference. IEEE SigPort. http://sigport.org/5453
Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini, 2020. Differentiable Branching in Deep Networks for Fast Inference. Available at: http://sigport.org/5453.
Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini. (2020). "Differentiable Branching in Deep Networks for Fast Inference." Web.
1. Simone Scardapane, Danilo Comminiello, Michele Scarpiniti, Enzo Baccarelli, Aurelio Uncini. Differentiable Branching in Deep Networks for Fast Inference [Internet]. IEEE SigPort; 2020. Available from : http://sigport.org/5453