PyTorchFI: A Runtime Perturbation Tool for DNNs

Mahmoud, Abdulrahman; Aggarwal, Neeraj; Nobbe, Alex; Vicarte, Jose Rodrigo Sanchez; Adve, Sarita V.; Fletcher, Christopher W.; Frosio, Iuri; Hari, Siva Kumar Sastry

doi:10.1109/dsn-w50199.2020.00014

Cited by 76 publications

(29 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The impact of faults in neural network, a core component in the learning-based navigation system, have been studied in a wide body of recent work. [19][20][21] build fault injection frameworks to quantify the error resilience of DNN applications. [22] evaluated the fault propagation of DNN focused on the vulnerability of different layers in an ASIC model.…”

Section: Related Workmentioning

confidence: 99%

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Wan¹,

Anwar²,

Hsiao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Learning-based navigation systems are widely used in autonomous applications, such as robotics, unmanned vehicles and drones. Specialized hardware accelerators have been proposed for highperformance and energy-efficiency for such navigational tasks. However, transient and permanent faults are increasing in hardware systems and can catastrophically violate tasks safety. Meanwhile, traditional redundancy-based protection methods are challenging to deploy on resource-constrained edge applications. In this paper, we experimentally evaluate the resilience of navigation systems with respect to algorithms, fault models and data types from both RL training and inference. We further propose two efficient fault mitigation techniques that achieve 2× success rate and 39% qualityof-flight improvement in learning-based navigation systems.

show abstract

Section: Related Workmentioning

confidence: 99%

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Wan¹,

Anwar²,

Hsiao³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In [14] the similarity of two consecutive video frames is exploited, expecting the prediction from the neural network to be the same for both. This simple fault-detection method doesn't incur overheads, however, the use cases for the technique is limited to video processing applications [15]. This paper presents two solutions for the detection of computational errors in neural inference deployed on an embedded platforms.…”

Section: Introductionmentioning

confidence: 99%

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

Safarpour

Deng

Massingham

et al. 2021

2021 IEEE Nordic Circuits and Systems Conference (NorCAS)

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 energy hungry building blocks in embedded systems. Without an error feedback mechanism, blind voltage down-scaling will result in degraded accuracy or total system failure. In this paper, solutions based on the inherent properties of Self-Supervised Learning (SSL) and Algorithm Based Fault Tolerance (ABFT) techniques were investigated. 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 lowoverhead 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 48%, without compromising the accuracy of the model was achieved.

show abstract

“…In [13] the similarity of two consecutive video frames is exploited, expecting the prediction from the neural network to be the same for both. This simple fault-detection method doesn't incur overheads, however, the use cases for the technique is limited to video processing applications [14]. This paper presents two solutions for the detection of computational errors in neural inference deployed on an embedded platforms.…”

Section: Introductionmentioning

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

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

PyTorchFI: A Runtime Perturbation Tool for DNNs

Cited by 76 publications

References 22 publications

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

Analyzing and Improving Fault Tolerance of Learning-Based Navigation Systems

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

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

Contact Info

Product

Resources

About