Resiliency of automotive object detection networks on GPU architectures

Lotfi, Atieh; Hukerikar, Saurabh; Balasubramanian, Keshav; Racunas, Paul; Saxena, N.R.; Bramley, Richard; Huang, Yuanding

doi:10.1109/itc44170.2019.9000150

Cited by 39 publications

(23 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a consequence, hardware-specific NNs, ranging from circuits for embedded machine learning applications to custom Very Large Scale Integration (VLSI) of neural networks in silicon, are traditionally considered tightly robust to hardware faults (HW faults). However, several recent studies have demonstrated that HW faults induced by an external perturbation (i.e., in a harsh environment) or due to silicon wearout and aging effects can significantly impact the inference leading to CNN prediction failures [10,11,12]. Therefore, ensuring the reliability of CNNs is crucial, especially when they are deployed in safety-critical and mission-critical applications, such as robotics, aeronautics, smart healthcare, and autonomous driving [13].…”

mentioning

confidence: 99%

Investigating data representation for efficient and reliable Convolutional Neural Networks

Ruospo

Sánchez

Traiola

et al. 2021

Microprocessors and Microsystems

View full text Add to dashboard Cite

mentioning

confidence: 99%

Investigating data representation for efficient and reliable Convolutional Neural Networks

Ruospo

Sánchez

Traiola

et al. 2021

Microprocessors and Microsystems

View full text Add to dashboard Cite

“…Concerning physical-based FI, we can cite [2], where the reliability dependence on three different Graphics Processing Unit (GPU) architectures (Kepler, Maxwell, and Pascal) was evaluated executing the Darknet Neural Network [8] when exposed to atmospheric-like neutrons. In [9], the authors analyse the reliability of a 54-layer Deep Neural Network injecting faults in the network weights and input data using an accelerated neutron beam for studying transient errors and FI tests to simulate permanent faults. The inferences target floating-point values, and the results show that object detection networks tend to generate wrong results when exposed to hardware faults.…”

Section: Related Workmentioning

confidence: 99%

Emulating the Effects of Radiation-Induced Soft-Errors for the Reliability Assessment of Neural Networks

Matanaluza

Ruospo

Söderström³

et al. 2022

IEEE Trans. Emerg. Topics Comput.

View full text Add to dashboard Cite

Convolutional Neural Networks (CNNs) are currently one of the most widely used predictive models in machine learning. Recent studies have demonstrated that hardware faults induced by radiation fields, including cosmic rays, may significantly impact the CNN inference leading to wrong predictions. Therefore, ensuring the reliability of CNNs is crucial, especially for safety-critical systems. In the literature, several works propose reliability assessments of CNNs mainly based on statistically injected faults. This work presents a software emulator capable of injecting real faults retrieved from radiation tests. Specifically, from the device characterisation of a DRAM memory, we extracted event rates and fault models. The software emulator can reproduce their incidence and access their effect on CNN applications with a reliability assessment precision close to the physical one. Radiation-based physical injections and emulator-based injections are performed on three CNNs (LeNet-5) exploiting different data representations. Their outcomes are compared, and the software results evidence that the emulator is able to reproduce the faulty behaviours observed during the radiation tests for the targeted CNNs. This approach leads to a more concise use of radiation experiments since the extracted fault models can be reused to explore different scenarios (e.g., impact on a different application).

show abstract

“…In [16], the authors studied the radiation-induced soft-errors based on AxC techniques applied to the data representation of weights in a Convolutional Neural Network (CNN). Withal, by using an accelerated neutron beam to inject transient errors and fault injection experiments for permanent errors, the reliability of a 54-layer CNN is assessed by exposing the entire GPU to the radiation source [17].…”

Section: A Radiation-based Fault Injectionmentioning

confidence: 99%

A Model-Based Framework to Assess the Reliability of Safety-Critical Applications

Luza

Ruospo

Bosio

et al. 2021

2021 24th International Symposium on Design and Diagnostics of Electronic Circuits &Amp; Systems (DDECS)

View full text Add to dashboard Cite

Solutions based on artificial intelligence and brain-inspired computations like Artificial Neural Networks (ANNs) are suited to deal with the growing computational complexity required by state-of-the-art electronic devices. Many applications that are being deployed using these computational models are considered safety-critical (e.g., self-driving cars), producing a pressing need to evaluate their reliability. Besides, state-of-the-art ANNs require significant memory resources to store their parameters (e.g., weights, activation values), which goes outside the possibility of many resourceconstrained embedded systems. In this light, Approximate Computing (AxC) has become a significant field of research to improve memory footprint, speed, and energy consumption in embedded and high-performance systems. The use of AxC can significantly reduce the cost of ANN implementations, but it may also reduce the inherent resiliency of this kind of application. On this scope, reliability assessments are carried out by performing fault injection test campaigns. The intent of the paper is to propose a framework that, relying on the results of radiation tests in Commercial-Off-The-Shelf (COTS) devices, is able to assess the reliability of a given application. To this end, a set of different radiation-induced errors in COTS memories is presented. Upon these, specific fault models are extracted to drive emulation-based fault injections.

show abstract

Resiliency of automotive object detection networks on GPU architectures

Cited by 39 publications

References 7 publications

Investigating data representation for efficient and reliable Convolutional Neural Networks

Investigating data representation for efficient and reliable Convolutional Neural Networks

Emulating the Effects of Radiation-Induced Soft-Errors for the Reliability Assessment of Neural Networks

A Model-Based Framework to Assess the Reliability of Safety-Critical Applications

Contact Info

Product

Resources

About