Algorithm-Based Fault Tolerance for Convolutional Neural Networks

Zhao, Kai; Di, Sheng; Li, Sihuan; Liang, Xin; Zhai, Yujia; Chen, Jieyang; Ouyang, Kaiming; Cappello, Franck

doi:10.1109/tpds.2020.3043449

Cited by 68 publications

(49 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This enables substantial power savings by protecting the correctness of arithmetic operations of the DNN. With the employed model the ABFT overheads were less than 8%, while it would be even less with larger networks [19]. The overheads in power consumption are in the same ball park, while the power saving greatly outweigh the overheads.…”

Section: B Error Detection Through Abftmentioning

confidence: 92%

See 1 more Smart Citation

Low-Voltage Energy Efficient Neural Inference by Leveraging Fault Detection Techniques

Safarpour¹,

Deng²,

Massingham³

et al. 2021

Preprint

View full text Add to dashboard Cite

Operating at reduced voltages offers substantial energy efficiency improvement but at the expense of increasing the probability of computational errors due to hardware faults. In this context, we targeted Deep Neural Networks (DNN) as emerging energy hungry building blocks in embedded applications. Without an error feedback mechanism, blind voltage down-scaling will result in degraded accuracy or total system failure. To enable safe voltage down-scaling, in this paper two solutions based on Self-Supervised Learning (SSL) and Algorithm Based Fault Tolerance (ABFT) were developed. A DNN model trained on MNIST data-set was deployed on a Field Programmable Gate Array (FPGA) that operated at reduced voltages and employed the proposed schemes. The SSL approach provides extremely low-overhead (≈0.2%) fault detection at the cost of lower error coverage and extra training, while ABFT incurs less than 8%overheads at run-time with close to 100% error detection rate. By using the solutions, substantial energy savings, i.e., up to 40.3%,without compromising the accuracy of the model was achieved

show abstract

Section: B Error Detection Through Abftmentioning

confidence: 92%

“…As shown in Fig. 3, the checksum of the output vector after each matrix-matrix multiplication or addition is inspected to verify the correctness of the operations [19].…”

Section: B Algorithm Based Fault Tolerancementioning

confidence: 99%

Low-Voltage Energy Efficient Neural Inference by Leveraging Fault Detection Techniques

Safarpour¹,

Deng²,

Massingham³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…When the testing process is not synchronized in time, then its performance is reduced, so it can be enhanced by a new perceptive of an algorithm for good repairing logic through proper synchronization of timing. Authors in [27] have proposed fault-tolerant convolutional neural networks (CNNs) for correcting serious errors in the real-time system.…”

Section: Literature Reviewmentioning

confidence: 99%

Adaptive Fault-Tolerant System and Optimal Power Allocation for Smart Vehicles in Smart Cities Using Controller Area Network

Biswal

Singh

Pattanayak

et al. 2021

Security and Communication Networks

View full text Add to dashboard Cite

Nowadays, the power consumption and dependable repeated data collection are causing the main issue for fault or collision in controller area network (CAN), which has a great impact for designing autonomous vehicle in smart cities. Whenever a smart vehicle is designed with several sensor nodes, Internet of Things (IoT) modules are linked through CAN for reliable transmission of a message for avoiding collision, but it is failed in communication due to delay and collision in communication of message frame from a source node to the destination. Generally, the emerging role of IoT and vehicles has undoubtedly brought a new path for tomorrow’s cities. The method proposed in this paper is used to gain fault-tolerant capability through Probabilistic Automatic Repeat Request (PARQ) and also Probabilistic Automatic Repeat Request (PARQ) with Fault Impact (PARQ-FI), in addition to providing optimal power allocation in CAN sensor nodes for enhancing the performance of the process and also significantly acting a role for making future smart cities. Several message frames are needed to be retransmitted on PARQ and fault impact (PARQ-FI) calculates the message with a response probability of each node.

show abstract

“…As ABFT in this case acts as an early warning, it can be used to control the operating point. Zhao et al [19] integrated ABFT in convolution operations of DNNs and showed it incurred up to only 8% overheads.…”

Section: Total Energy Optimizationmentioning

confidence: 99%

A High-Level Approach for Energy Efficiency Improvement of FPGAs by Voltage Trimming

Safarpour¹

2021

Preprint

View full text Add to dashboard Cite

show abstract

Algorithm-Based Fault Tolerance for Convolutional Neural Networks

Cited by 68 publications

References 39 publications

Low-Voltage Energy Efficient Neural Inference by Leveraging Fault Detection Techniques

Low-Voltage Energy Efficient Neural Inference by Leveraging Fault Detection Techniques

Adaptive Fault-Tolerant System and Optimal Power Allocation for Smart Vehicles in Smart Cities Using Controller Area Network

A High-Level Approach for Energy Efficiency Improvement of FPGAs by Voltage Trimming

Contact Info

Product

Resources

About