Fault Injection for TensorFlow Applications

Niranjhana Narayanan; Zitao Chen; Bo Fang; Guanpeng Li; Karthik Pattabiraman; Nathan Debardeleben

doi:10.1109/TDSC.2022.3175930

Back

Journal article

Fault Injection for TensorFlow Applications

Niranjhana Narayanan, Zitao Chen, Bo Fang, Guanpeng Li, Karthik Pattabiraman and Nathan Debardeleben

IEEE transactions on dependable and secure computing, Vol.20(4), pp.2677-2695

07/2023

DOI: 10.1109/TDSC.2022.3175930

Files and links (1)

url

https://www.osti.gov/servlets/purl/1994506View

Open Access

Abstract

As machine learning (ML) has seen increasing adoption in safety-critical domains (e.g., autonomous vehicles), the reliability of ML systems has also grown in importance. While prior studies have proposed techniques to enable efficient error-resilience (e.g., selective instruction duplication), a fundamental requirement for realizing these techniques is a detailed understanding of the application's resilience. In this work, we present TensorFI 1 and TensorFI 2, high-level fault injection (FI) frameworks for TensorFlow-based applications. TensorFI 1 and 2 are able to inject both hardware and software faults in any general TensorFlow 1 and 2 program respectively. Both are configurable FI tools that are flexible, easy to use, and portable. They can be integrated into existing TensorFlow programs to assess their resilience for different fault types (e.g., bit-flips in particular operations or layers). We use the TensorFI 1 and TensorFI 2 to evaluate the resilience of 11 and 10 ML programs written in TensorFlow, including DNNs used in the autonomous vehicle domain. The results give us insights into why some of the models are more resilient. We also measure the performance overheads of the two injectors, and present 4 case studies, two for each tool, to demonstrate their utility.

Machine Learning

Safety

Software

Computational modeling

Data models

Deep Neural Networks

Error Resilience

Fault Injection

Hardware

Reliability

Resilience

TensorFlow

Details

Title: Subtitle: Fault Injection for TensorFlow Applications
Creators: Niranjhana Narayanan - University of British Columbia
Zitao Chen - University of British Columbia
Bo Fang - Pacific Northwest National Laboratory
Guanpeng Li - University of Iowa
Karthik Pattabiraman - University of British Columbia
Nathan Debardeleben - Los Alamos National Laboratory
Resource Type: Journal article
Publication Details: IEEE transactions on dependable and secure computing, Vol.20(4), pp.2677-2695
DOI: 10.1109/TDSC.2022.3175930
ISSN: 1545-5971
eISSN: 1941-0018
Publisher: IEEE
Grant note: Canadian Network for Research and Innovation in Machining Technology Natural Sciences and Engineering Research Council of Canada (10.13039/501100002790) Huawei Technologies (10.13039/501100003816)
Language: English
Electronic publication date: 07/15/2022
Date published: 07/2023
Academic Unit: Computer Science
Record Identifier: 9984411089402771

Metrics

20 Record Views

19 Times Cited - Web of Science