Towards a robust and trustworthy machine learning system development: An engineering perspective

Xiong, Pulei; Buffett, Scott; Iqbal, Shahrear; Lamontagne, Philippe; Mamun, Mohammad Saiful Islam; Molyneaux, Heather

doi:10.1016/j.jisa.2022.103121

Cited by 16 publications

(9 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These models consider dynamic factors, model map information, and can incorporate interactionrelated factors, allowing adaptation to complex scenes [1]. However, many deep learning models are known to be sensitive to small errors and susceptible to external attacks, potentially resulting in undesirable behavior and decreased performance [13]. Therefore, ensuring reliable performance (often called robustness) is key to the safe deployment of these deep learning models.…”

Section: Introductionmentioning

confidence: 99%

On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles

Zhang

Sun

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Autonomous vehicles rely on accurate trajectory prediction to inform decision-making processes related to navigation and collision avoidance. However, current trajectory prediction models show signs of overfitting, which may lead to unsafe or suboptimal behavior. To address these challenges, this paper presents a comprehensive framework that categorizes and assesses the definitions and strategies used in the literature on evaluating and improving the robustness of trajectory prediction models. This involves a detailed exploration of various approaches, including data slicing methods, perturbation techniques, model architecture changes, and post-training adjustments. In the literature, we see many promising methods for increasing robustness, which are necessary for safe and reliable autonomous driving.

show abstract

Section: Introductionmentioning

confidence: 99%

On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles

Zhang

Sun

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

“…Social engineering attacks can involve the creation of fake RTU-CAV messages, phishing, or GPS map poisoning. Social engineering attacks [152] Data sanitization techniques [153], [154], [155], Adversarial sample thwarting (e.g., data transformation, noise filtering, mapping to normal samples) [156], Generalization enhancement (e.g., bagging, random subspace method, antidote, etc.) [151], Training data filtering (e.g., input manipulation detection, gradient shaping) [46], Robust learning (e.g., model robustifying and verification) [46], Feature obfuscation [157], Active learning systems [157], Min-max optimization [158], [159] Gradient-based approach [160], Enhance generalization capability using adversarial feature selection method [161].…”

Section: Defenses Against Social Engineering Attackmentioning

confidence: 99%

An investigation of cyber-attacks and security mechanisms for connected and autonomous vehicles

Gupta¹,

Maple²

2023

Preprint

View full text Add to dashboard Cite

<p>Edge Computing-based Connected Autonomous Vehicles (EC-CAVs) can fulfill the emerging demand for smart transportation on a global scale. Such innovations to transportation ecosystems can bring manyfold benefits from fully autonomous driving services to proactive vehicle monitoring and traffic management. Given the complexity involved in the deployment of EC-CAVs, zero-tolerance safety and security measures must be incorporated to avert vehicle immobilization, road accidents, disclosure of sensitive data, or any potential threats. This paper presents a reference architecture of EC-CAVs to investigate existing and emerging cyber-threats for deriving a common attack taxonomy for the EC-CAVs ecosystem. Subsequently, a comprehensive survey of security mechanisms is presented that can ensure safe and secure transportation of passengers from one destination to another using an EC-CAVs ecosystem.</p>

show abstract

“…Machine learning has evolved over several decades and there are many classifications of the techniques in the literature (Ahmad et al, 2021; Xiong et al, 2022). One of such general machine learning classifications is reported by Umenweke et al (2022).…”

Section: Overview Of Orc Plant and Data‐driven Modeling Approachmentioning

confidence: 99%

Machine learning for design and optimization of organic Rankine cycle plants: A review of current status and future perspectives

Oyekale

Oreko

2023

WIREs Energy & Environment

View full text Add to dashboard Cite

The organic Rankine cycle (ORC) is widely acknowledged as a sustainable power cycle. However, the traditional approach commonly adopted for its optimal design involves sequential consideration of working fluid selection, plant configuration, and component types, before the optimization of state parameters. This way, the design process fails to achieve an optimal design in most cases, since the process relies heavily on empirical judgments. To improve the design process, researchers have been exploring lately the suitability of machine learning techniques. It is however not clear yet if data‐driven designs of ORC plants are practically viable and accurate. To bridge this gap, this article reviews literature studies in the field. Overviews were first presented on the ORC technology and machine learning modeling approaches. Next, studies that applied machine‐learning methods for the design and performance prediction of ORC plants were discussed. Furthermore, studies that focused on ORC machine learning optimizations were discussed. The artificial neural network (ANN) approach was observed as the technique most frequently applied for ORC design and optimization. Additionally, researchers agree in general that machine‐learning methods can achieve accurate results, with significant reductions of computational time and cost. However, there is the risk of using inadequate data size in the machine learning design approach, or insufficient data set training time, all of which can affect accuracy. It is hoped that this effort would spur the practical implementation of machine learning techniques in the future design and optimization of ORC plants, toward the achievement of more sustainable energy technology. This article is categorized under: Sustainable Energy > Energy Efficiency Sustainable Energy > Other Renewables

show abstract

Towards a robust and trustworthy machine learning system development: An engineering perspective

Cited by 16 publications

References 55 publications

On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles

On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles

An investigation of cyber-attacks and security mechanisms for connected and autonomous vehicles

Machine learning for design and optimization of organic Rankine cycle plants: A review of current status and future perspectives

Contact Info

Product

Resources

About