PerforatedCNNs: Acceleration through Elimination of Redundant Convolutions
Authors:
Michael Figurnov,
Aijan Ibraimova,
Dmitry Vetrov,
Pushmeet Kohli
Abstract:
We propose a novel approach to reduce the computational cost of evaluation of convolutional neural networks, a factor that has hindered their deployment in low-power devices such as mobile phones. Inspired by the loop perforation technique from source code optimization, we speed up the bottleneck convolutional layers by skip** their evaluation in some of the spatial positions. We propose and ana…
▽ More
We propose a novel approach to reduce the computational cost of evaluation of convolutional neural networks, a factor that has hindered their deployment in low-power devices such as mobile phones. Inspired by the loop perforation technique from source code optimization, we speed up the bottleneck convolutional layers by skip** their evaluation in some of the spatial positions. We propose and analyze several strategies of choosing these positions. We demonstrate that perforation can accelerate modern convolutional networks such as AlexNet and VGG-16 by a factor of 2x - 4x. Additionally, we show that perforation is complementary to the recently proposed acceleration method of Zhang et al.
△ Less
Submitted 15 October, 2016; v1 submitted 30 April, 2015;
originally announced April 2015.
Using cascade development universality for thin calorimeter
Authors:
A. Sh. Gaitinov,
S. A. Ibraimova,
I. A. LebedeV,
A. I Lebedeva
Abstract:
A method for measurement of energy of high-energy particles by a thin calorimeter, is presented. The method is based on the correlation analysis of dependence of number of secondary particles, $N_e$, at observation level and the relation of number of particles, $dN$, at two levels, divided by an absorber layer. It is shown, that use of correlation curves ($log N_e$ versus $dN$) allows to reduce es…
▽ More
A method for measurement of energy of high-energy particles by a thin calorimeter, is presented. The method is based on the correlation analysis of dependence of number of secondary particles, $N_e$, at observation level and the relation of number of particles, $dN$, at two levels, divided by an absorber layer. It is shown, that use of correlation curves ($log N_e$ versus $dN$) allows to reduce essentially errors of definition of energy of the primary particle, which are connected with uncertainty of a primary nucleus and with fluctuations in development of cascade process.
△ Less
Submitted 20 October, 2013;
originally announced October 2013.
