Thales: Formulating and Estimating Architectural Vulnerability Factors for DNN Accelerators

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tyagi, Abhishek, Gan, Yiming, Liu, Shaoshan, Yu, Bo, Whatmough, Paul, Zhu, Yuhao
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929201543643136
author Tyagi, Abhishek
Gan, Yiming
Liu, Shaoshan
Yu, Bo
Whatmough, Paul
Zhu, Yuhao
author_facet Tyagi, Abhishek
Gan, Yiming
Liu, Shaoshan
Yu, Bo
Whatmough, Paul
Zhu, Yuhao
contents As Deep Neural Networks (DNNs) are increasingly deployed in safety critical and privacy sensitive applications such as autonomous driving and biometric authentication, it is critical to understand the fault-tolerance nature of DNNs. Prior work primarily focuses on metrics such as Failures In Time (FIT) rate and the Silent Data Corruption (SDC) rate, which quantify how often a device fails. Instead, this paper focuses on quantifying the DNN accuracy given that a transient error has occurred, which tells us how well a network behaves when a transient error occurs. We call this metric Resiliency Accuracy (RA). We show that existing RA formulation is fundamentally inaccurate, because it incorrectly assumes that software variables (model weights/activations) have equal faulty probability under hardware transient faults. We present an algorithm that captures the faulty probabilities of DNN variables under transient faults and, thus, provides correct RA estimations validated by hardware. To accelerate RA estimation, we reformulate RA calculation as a Monte Carlo integration problem, and solve it using importance sampling driven by DNN specific heuristics. Using our lightweight RA estimation method, we show that transient faults lead to far greater accuracy degradation than what todays DNN resiliency tools estimate. We show how our RA estimation tool can help design more resilient DNNs by integrating it with a Network Architecture Search framework.
format Preprint
id arxiv_https___arxiv_org_abs_2212_02649
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Thales: Formulating and Estimating Architectural Vulnerability Factors for DNN Accelerators
Tyagi, Abhishek
Gan, Yiming
Liu, Shaoshan
Yu, Bo
Whatmough, Paul
Zhu, Yuhao
Hardware Architecture
Cryptography and Security
Machine Learning
As Deep Neural Networks (DNNs) are increasingly deployed in safety critical and privacy sensitive applications such as autonomous driving and biometric authentication, it is critical to understand the fault-tolerance nature of DNNs. Prior work primarily focuses on metrics such as Failures In Time (FIT) rate and the Silent Data Corruption (SDC) rate, which quantify how often a device fails. Instead, this paper focuses on quantifying the DNN accuracy given that a transient error has occurred, which tells us how well a network behaves when a transient error occurs. We call this metric Resiliency Accuracy (RA). We show that existing RA formulation is fundamentally inaccurate, because it incorrectly assumes that software variables (model weights/activations) have equal faulty probability under hardware transient faults. We present an algorithm that captures the faulty probabilities of DNN variables under transient faults and, thus, provides correct RA estimations validated by hardware. To accelerate RA estimation, we reformulate RA calculation as a Monte Carlo integration problem, and solve it using importance sampling driven by DNN specific heuristics. Using our lightweight RA estimation method, we show that transient faults lead to far greater accuracy degradation than what todays DNN resiliency tools estimate. We show how our RA estimation tool can help design more resilient DNNs by integrating it with a Network Architecture Search framework.
title Thales: Formulating and Estimating Architectural Vulnerability Factors for DNN Accelerators
topic Hardware Architecture
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2212.02649