Training Modern Deep Neural Networks for Memory-Fault Robustness
Authors:
- Ghouthi Boukli HaceneGHGhouthi Boukli Hacene
- 1 May 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21581525,p. 1-5
Abstract
Because deep neural networks (DNNs) rely on a large number of parameters and computations, their implementation in energy-constrained systems is challenging. In this paper, we investigate the solution of reducing the supply voltage of the memories used in the system, which results in bit-cell faults. We explore the robustness of state-of-the-art DNN architectures towards such defects and propose a regularizer meant to mitigate their effects on accuracy. Our experiments clearly demonstrate the interest of operating the system in a faulty regime to save energy without reducing accuracy.Keywords
All Related Versions
This publication has 16 references indexed in Scilit:
- 14.3 A 28nm SoC with a 1.2GHz 568nJ/prediction sparse deep-neural-network engine with >0.1 timing error rate tolerance for IoT applications2017
- Variation-Tolerant Architectures for Convolutional Neural Networks in the Near Threshold Voltage Regime2016
- Compression of Deep Neural Networks on the FlyLecture Notes in Computer Science, 2016
- MinervaACM SIGARCH Computer Architecture News, 2016
- Deep Residual Learning for Image Recognition2016
- Deep learningNature, 2015
- DaDianNao: A Machine-Learning Supercomputer2014
- Overcoming Computational Errors in Sensing Platforms Through Embedded Machine-Learning KernelsIEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2014
- A defect-tolerant accelerator for emerging high-performance applications2012
- Near-Threshold Computing: Reclaiming Moore's Law Through Energy Efficient Integrated CircuitsProceedings of the IEEE, 2010