Survey of Attacks and Defenses on Edge-Deployed Neural Networks

Isakov, Mihailo; Gadepally, Vijay; Gettings, Karen; Kinsy, Michel A.

doi:10.1109/hpec.2019.8916519

Cited by 30 publications

(26 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The security of DNNs on embedded devices has been discussed recently [IGGK19,SW19]. Several of the attacks against DNNs of embedded devices (e.g., against cryptographic devices [Koc96,KJJ99]) have been severe because the attackers easily reverseengineered them or measured the side-channel leaks, such as processing time and power consumption.…”

Section: Introductionmentioning

confidence: 99%

Timing Black-Box Attacks: Crafting Adversarial Examples through Timing Leaks against DNNs on Embedded Devices

Nakai

Suzuki

Fujino

2021

TCHES

View full text Add to dashboard Cite

Deep neural networks (DNNs) have been applied to various industries. In particular, DNNs on embedded devices have attracted considerable interest because they allow real-time and distributed processing on site. However, adversarial examples (AEs), which add small perturbations to the input data of DNNs to cause misclassification, are serious threats to DNNs. In this paper, a novel black-box attack is proposed to craft AEs based only on processing time, i.e., the side-channel leaks from DNNs on embedded devices. Unlike several existing black-box attacks that utilize output probability, the proposed attack exploits the relationship between the number of activated nodes and processing time without using training data, model architecture, parameters, substitute models, or output probability. The perturbations for AEs are determined by the differential processing time based on the input data of the DNNs in the proposed attack. The experimental results show that the AEs of the proposed attack effectively cause an increase in the number of activated nodes and the misclassification of one of the incorrect labels against the DNNs on a microcontroller unit. Moreover, these results indicate that the attack can evade gradient-masking and confidence reduction countermeasures, which conceal the output probability, to prevent the crafting of AEs against several black-box attacks. Finally, the countermeasures against the attack are implemented and evaluated to clarify that the implementation of an activation function with data-dependent timing leaks is the cause of the proposed attack.

show abstract

Section: Introductionmentioning

confidence: 99%

Timing Black-Box Attacks: Crafting Adversarial Examples through Timing Leaks against DNNs on Embedded Devices

Nakai

Suzuki

Fujino

2021

TCHES

View full text Add to dashboard Cite

show abstract

“…Another approach is Edge Intelligence (EdgeAI) where the AI model is distributed across network edges. Several works have discussed the convergence of edge and AI [ 30 , 31 , 32 , 33 , 34 , 35 ]. AI model can be pre-trained then modified and optimized to be appropriate to run in the resource-constrained edges.…”

Section: Background and Related Workmentioning

confidence: 99%

Distributed Artificial Intelligence-as-a-Service (DAIaaS) for Smarter IoE and 6G Environments

Janbi

Katib

Albeshri

et al. 2020

Sensors

View full text Add to dashboard Cite

Artificial intelligence (AI) has taken us by storm, helping us to make decisions in everything we do, even in finding our “true love” and the “significant other”. While 5G promises us high-speed mobile internet, 6G pledges to support ubiquitous AI services through next-generation softwarization, heterogeneity, and configurability of networks. The work on 6G is in its infancy and requires the community to conceptualize and develop its design, implementation, deployment, and use cases. Towards this end, this paper proposes a framework for Distributed AI as a Service (DAIaaS) provisioning for Internet of Everything (IoE) and 6G environments. The AI service is “distributed” because the actual training and inference computations are divided into smaller, concurrent, computations suited to the level and capacity of resources available with cloud, fog, and edge layers. Multiple DAIaaS provisioning configurations for distributed training and inference are proposed to investigate the design choices and performance bottlenecks of DAIaaS. Specifically, we have developed three case studies (e.g., smart airport) with eight scenarios (e.g., federated learning) comprising nine applications and AI delivery models (smart surveillance, etc.) and 50 distinct sensor and software modules (e.g., object tracker). The evaluation of the case studies and the DAIaaS framework is reported in terms of end-to-end delay, network usage, energy consumption, and financial savings with recommendations to achieve higher performance. DAIaaS will facilitate standardization of distributed AI provisioning, allow developers to focus on the domain-specific details without worrying about distributed training and inference, and help systemize the mass-production of technologies for smarter environments.

show abstract

“…They provide a model for classifying attacks against online machine learning algorithms. Most of the studies in these fields have been focused on specific adversarial attacks and generally, presented the theoretical discussion of adversarial machine learning area [23,25].…”

Section: Related Workmentioning

confidence: 99%

A Generative Model based Adversarial Security of Deep Learning and Linear Classifier Models

Sivaslioglu

Çatak

ŞAHİNBAŞ

2021

IJCAI

View full text Add to dashboard Cite

Survey of Attacks and Defenses on Edge-Deployed Neural Networks

Cited by 30 publications

References 47 publications

Timing Black-Box Attacks: Crafting Adversarial Examples through Timing Leaks against DNNs on Embedded Devices

Timing Black-Box Attacks: Crafting Adversarial Examples through Timing Leaks against DNNs on Embedded Devices

Distributed Artificial Intelligence-as-a-Service (DAIaaS) for Smarter IoE and 6G Environments

A Generative Model based Adversarial Security of Deep Learning and Linear Classifier Models

Contact Info

Product

Resources

About