Impact of System Resources on Performance of Deep Neural Network
Parijat Dube, Zehra Sura
ISPASS 2018
The purpose of this paper is to study the loss probabilities of messages in an M/M/1/K queueing system where in addition to losses due to buffer overflow there are also random losses in the incoming and outgoing links. We focus on the influence of adding redundant packets to the messages (as in error correction coding, e.g. Reed-Solomon code, etc.). In the first part we use multi-dimensional probability generating functions for solving the recursions which generalize those introduced by Cidon et al. [IEEE Trans. Inform. Theory 39 (1) (1993) 98] for computing the loss probabilities and derive analytical formulae for a special case. In the second part of the paper we use combinatorial arguments and Ballot theorem results to alternatively obtain the loss probabilities. The analytical results allow us to investigate when does adding redundancy decrease the loss probabilities. © 2003 Elsevier Science B.V. All rights reserved.
Parijat Dube, Zehra Sura
ISPASS 2018
Paula Austel, Han Chen, et al.
CF 2015
Sanghamitra Dutta, Gauri Joshi, et al.
AISTATS 2018
Parijat Dube, Eitan Altman
IEEE INFOCOM 2003