Towards a Bayesian Approach for Assessing Fault Tolerance of Deep Neural Networks
Subho S. Banerjee, James Cyriac, Saurabh Jha, Zbigniew T. Kalbarczyk, and Ravishankar K. Iyer
DSN 2019
Abstract
This paper presents Bayesian Deep Learning based Fault Injection (BDLFI), a novel methodology for fault injection in neural networks (NNs) or more generally differentiable programs. BDLFI uses (1) Bayesian Deep Learning to model the propagation of faults, and (2) Markov Chain Monte Carlo inference to quantify the effect of faults on the outputs of a NN. We demonstrate BDLFI on two representative networks and find that our results challenge pre-existing results in the field.
Citation
@INPROCEEDINGS{Banerjee2019_DSN,
author={S. S. {Banerjee} and J. {Cyriac} and S. {Jha} and Z. T. {Kalbarczyk} and R. K. {Iyer}},
booktitle={2019 49th Annual IEEE/IFIP International Conference on Dependable Systems and Networks – Supplemental Volume (DSN-S)},
title={Towards a Bayesian Approach for Assessing Fault Tolerance of Deep Neural Networks},
year={2019},
volume={},
number={},
pages={25-26},
keywords={Fault Injection;Neural Networks},
doi={10.1109/DSN-S.2019.00018},
ISSN={},
month={June},
}