Analyzing and Increasing the Reliability of Convolutional Neural Networks on GPUs
The DOI matches its Crossref record; 68 of the 68 references it cites were found in a registry or an index.
Abstract
Graphics processing units (GPUs) are playing a critical role in convolutional neural networks (CNNs) for image detection. As GPU-enabled CNNs move into safety-critical environments, reliability is becoming a growing concern. In this paper, we evaluate and propose strategies to improve the reliability of object detection algorithms, as run on three NVIDIA GPU architectures. We consider three algorithms: 1) you only look once; 2) a faster region-based CNN (Faster R-CNN); and 3) a residual network, exposing live hardware to neutron beams. We complement our beam experiments with fault injection to better characterize fault propagation in CNNs. We show that a single fault occurring in a GPU tends to propagate to multiple active threads, significantly reducing the reliability of a CNN. Moreover, relying on error correcting codes dramatically reduces the number of silent data corruptions (SDCs), but does not reduce the number of critical errors (i.e., errors that could potentially impact safety-critical applications). Based on observations on how faults propagate on GPU architectures, we propose effective strategies to improve CNN reliability. We also consider the benefits of using an algorithm-based fault-tolerance technique for matrix multiplication, which can correct more than 87% of the critical SDCs in a CNN, while redesigning maxpool layers of the CNN to detect up to 98% of critical SDCs.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“as commented in [9], SASSIFI is the only tool working with the NVIDIA proprietary libraries generally employed when implementing CNNs.”
From Fast and Accurate Error Simulation for CNNs Against Soft Errors (2023) · cites this paper as [9]
In the citing paper
“radiation test offers the most adherent solution to cause soft errors, however, apart from the highly expensive equipment and the complex setup, there is little to no controllability on the injection process, and the observability is only on the primary outputs. As a result, it is suitable for final validation and blackbox campaigns, not for in-depth analysis and investigations [9], [36].”
From Fast and Accurate Error Simulation for CNNs Against Soft Errors (2023) · cites this paper as [9]
Citations
Cites
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 1×
“Specifically, neural networks for object detection have the peculiarity of using convolution to reduce the information to be processed [15], a class of computations where reliability has not been studied deeply to date.”
BackgroundYou Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803
An introduction to ROC analysis
Fawcett, T. (2006) · Pattern Recognition Letters · doi:10.1016/j.patrec.2005.10.010 · Cited by 21,879 · cited 2×
“5 to compare the corrupted output of a CNN with the radiation-free output, following the T J tradeoff discussion presented in Fawcett et al. [47], which is valid, independent of the source of detection imprecision (intrinsic algorithm detection imprecision or radiation-induced corruption).”
MethodThe pascal visual object classes (VOC) challenge
Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010) · International Journal of Computer Vision · doi:10.1007/s11263-009-0275-4 · Cited by 19,599 · cited 1×
“We use the PASCAL VOC 2012 [48] and the Caltech Pedes-trian datasets [49] as input for our study.”
MethodCaffe: Convolutional Architecture for Fast Feature Embedding
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., & Darrell, T. (2014) · Proceedings of the 22nd ACM international conference on Multimedia · doi:10.1145/2647868.2654889 · Cited by 11,077 · cited 2×
“Convolutional neural networks (CNNs) have, in fact, been deployed in self-driving cars to dynamically identify/classify objects [2], [3].”
BackgroundMethodIdentity mappings in deep residual networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46493-0_38 · Cited by 10,741
Identity Mappings in Deep Residual Networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · arXiv · doi:10.48550/arxiv.1603.05027 · Cited by 10,665 · cited 2×
“This process is performed repeatedly for all layers in the network, allowing us to use deeper network structures (e.g., 50 to 1000 layers) without impacting precision, and without increasing the algorithmic complexity [45].”
MethodIn-Datacenter Performance Analysis of a Tensor Processing Unit
Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., Bates, S., Bhatia, S., Boden, N., Borchers, A., et al. (2017) · Proceedings of the 44th Annual International Symposium on Computer Architecture · doi:10.1145/3079856.3080246 · Cited by 4,537 · cited 1×
“As GPUs are highly efficient for accelerating matrix multiplication, they are a great target for CNNs. Specific processors can be designed for DNN, achieving good performance and low power consumption [42], [43].”
Background
Research works printed but not linked: 4
- sridharan, 2009, The effect of input data on program vulnerability, Workshop on Silicon Errors in Logic-System Effects, 1
- 10.1109/TNS.2014.2301768 · DOI
- 2018, Log database for IEEE transaction on reliability, Github
- collobert, 0, BigLearn NIPS Workshop
Cited by
A survey on modeling and improving reliability of DNN algorithms and accelerators
Mittal, S. (2020) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2019.101689 · Cited by 175
Reliable Data Transmission Model for Mobile Ad Hoc Network Using Signcryption Technique
Elhoseny, M., & Shankar, K. (2020) · IEEE Transactions on Reliability · doi:10.1109/tr.2019.2915800 · Cited by 159
“Steering overhead is the principle issue with course disclosure, which causes data loss [19].”
BackgroundA Low-cost Fault Corrector for Deep Neural Networks through Range Restriction
Chen, Z., Li, G., & Pattabiraman, K. (2021) · 2021 51st Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN) · doi:10.1109/dsn48987.2021.00018 · Cited by 104
“Techniques based on Algorithm-based Fault Tolerance (ABFT) have also been proposed for detecting faults in particular layers in DNNs (e.g., Conv layer [17], [18], [61]; matrix multiplication [62]).”
Soft errors in DNN accelerators: A comprehensive review
Ibrahim, Y., Wang, H., Liu, J., Wei, J., Chen, L., Rech, P., Adam, K., & Guo, G. (2020) · Microelectronics Reliability · doi:10.1016/j.microrel.2020.113969 · Cited by 102
Self-citationA Systematic Literature Review on Hardware Reliability Assessment Methods for Deep Neural Networks
Ahmadilivani, M. H., Taheri, M., Raik, J., Daneshtalab, M., & Jenihhin, M. (2024) · ACM Computing Surveys · doi:10.1145/3638242 · Cited by 134
“The general model for classifying faults in the mentioned works is as follows: — Masked: Fault does not affect the output, — SDC: Output confidence score differs from that of the golden model, — DUE: The program hangs or the system reboots (also called Crash in References [10, 121]).”
BackgroundMethodInfluential
Metadata
- Authors
- Santos, F. F. D., Pimenta, P. F., Lunardi, C., Draghetti, L., Carro, L., Kaeli, D., & Rech, P.
- Affiliations
- Universidade Federal do Rio Grande do Sul (BR)
- Northeastern University (US)
- ORCID
- Venue
- IEEE Transactions on Reliability
- Published
- 2019-06
- Volume
- 68
- Issue
- 2
- Pages
- 663-677
- ISSN
- 0018-9529 (ISSN-L)
- 1558-1721
- OpenAlex
- W2901848761
- Cited by
- 226
- References
- 67
- Citations by year
- 2026: 18 · 2025: 18 · 2024: 41 · 2023: 33 · 2022: 42 · 2021: 34 · 2020: 27 · 2019: 13
- FWCI
- 7.08
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Transactions on Reliability (publishedVersion)
- Institutional Research Information System (Università degli Studi di Trento) (submittedVersion)
- Science and Technology Facilities Council (submittedVersion)
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Radiation Effects in Electronics
- Physical Sciences › Computer Science › Computer Networks and Communications › Distributed systems and fault tolerance
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Keywords
- convolutional neural networks, GPU reliability, fault injection, silent data corruption, error-correcting codes, algorithm-based fault tolerance, matrix multiplication, object detection, safety-critical systems, fault propagation, NVIDIA GPUs
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Analyzing and Increasing the Reliability of Convolutional Neural Networks on GPUs. https://app.cochat.ai/papers/10.1109/tr.2018.2878387
- The paper
- Santos, F. F. D., Pimenta, P. F., Lunardi, C., Draghetti, L., Carro, L., Kaeli, D., & Rech, P. (2019). Analyzing and Increasing the Reliability of Convolutional Neural Networks on GPUs. IEEE Transactions on Reliability. https://doi.org/10.1109/tr.2018.2878387
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.