Adversarial Examples in Deep Neural Networks: An Overview

Balda, Emilio Rafael; Behboodi, Arash; Mathar, Rudolf

doi:10.1007/978-3-030-31760-7_2

Cited by 13 publications

(11 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another aspect of AI technical robustness that has been the focus of recent research is the field of adversarial attacks. A comprehensive review of these methods can be found in [60] [61] [62] [63] [64]. The aim of the researchers is to introduce in their models' layers of robustness such that the models are not misled by out of distribution examples, known or unknown attacks and targeted or untargeted attacks.…”

Section: B Technical Robustness and Learning Assurancementioning

confidence: 99%

An Explainable Artificial Intelligence (xAI) Framework for Improving Trust in Automated ATM Tools

Hernandez

Ayo

Panagiotakopoulos

2021

2021 IEEE/AIAA 40th Digital Avionics Systems Conference (DASC)

View full text Add to dashboard Cite

With the increased use of intelligent Decision Support Tools in Air Traffic Management (ATM) and inclusion of non-traditional entities, regulators and end users need assurance that new technologies such as Artificial Intelligence (AI) and Machine Learning (ML) are trustworthy and safe. Although there is a wide amount of research on the technologies themselves, there seem to be a gap between research projects and practical implementation due to different regulatory and practical challenges including the need for transparency and explainability of solutions. In order to help address these challenges, a novel framework to enable trust on AI-based automated solutions is presented based on current guidelines and end user feedback. Finally, recommendations are provided to bridge the gap between research and implementation of AI and ML-based solutions using our framework as a mechanism to aid advances of AI technology within ATM.

show abstract

Section: B Technical Robustness and Learning Assurancementioning

confidence: 99%

An Explainable Artificial Intelligence (xAI) Framework for Improving Trust in Automated ATM Tools

Hernandez

Ayo

Panagiotakopoulos

2021

2021 IEEE/AIAA 40th Digital Avionics Systems Conference (DASC)

View full text Add to dashboard Cite

show abstract

“…More recently, proposals have emerged based on training an ML model for a task and then first using that model to classify, then asking humans if that model's confidence is not high enough (Callaghan et al 2018). The effectiveness of such an approach is, consequently, heavily dependent on the reliability of machine confidence, which has shown to be very poor especially for deep learning (Guo et al 2017;Balda, Behboodi, and Mathar 2020).…”

Section: Ai Workflows and The Metrics That Mattermentioning

confidence: 99%

The Science of Rejection: A Research Area for Human Computation

Sayin¹,

Yang²,

Passerini³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In general, AI models are trained on the natural distribution of the data considered in the specific problem (e.g., the distribution of traffic sign images). This distribution, however, lies on a very low-dimensional manifold as compared to the complete input space (e.g., all possible images of the same resolution) (Tanay and Griffin, 2016 ; Balda et al, 2020 ), which is sometimes referred to as the “curse of dimensionality.” Table 1 shows that the size of the input space for some common tasks is extremely large. Even rather simple and academic AI models as e.g., LeNet-5 for handwritten digit recognition have a huge input space.…”

Section: Key Factors Underlying Ai-specific Vulnerabilitiesmentioning

confidence: 99%

“…Besides the arms race in practical attacks and defenses, adversarial attacks have also sparked interest from a theoretical perspective (Goodfellow et al, 2015 ; Tanay and Griffin, 2016 ; Biggio and Roli, 2018 ; Khoury and Hadfield-Menell, 2018 ; Madry et al, 2018 ; Ilyas et al, 2019 ; Balda et al, 2020 ). Several publications deal with their essential characteristics.…”

Section: Key Factors Underlying Ai-specific Vulnerabilitiesmentioning

confidence: 99%

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

2020

View full text Add to dashboard Cite

This article deals with the IT security of connectionist artificial intelligence (AI) applications, focusing on threats to integrity, one of the three IT security goals. Such threats are for instance most relevant in prominent AI computer vision applications. In order to present a holistic view on the IT security goal integrity, many additional aspects, such as interpretability, robustness and documentation are taken into account. A comprehensive list of threats and possible mitigations is presented by reviewing the state-of-the-art literature. AI-specific vulnerabilities, such as adversarial attacks and poisoning attacks are discussed in detail, together with key factors underlying them. Additionally and in contrast to former reviews, the whole AI life cycle is analyzed with respect to vulnerabilities, including the planning, data acquisition, training, evaluation and operation phases. The discussion of mitigations is likewise not restricted to the level of the AI system itself but rather advocates viewing AI systems in the context of their life cycles and their embeddings in larger IT infrastructures and hardware devices. Based on this and the observation that adaptive attackers may circumvent any single published AI-specific defense to date, the article concludes that single protective measures are not sufficient but rather multiple measures on different levels have to be combined to achieve a minimum level of IT security for AI applications.

show abstract

Adversarial Examples in Deep Neural Networks: An Overview

Cited by 13 publications

References 29 publications

An Explainable Artificial Intelligence (xAI) Framework for Improving Trust in Automated ATM Tools

An Explainable Artificial Intelligence (xAI) Framework for Improving Trust in Automated ATM Tools

The Science of Rejection: A Research Area for Human Computation

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

Contact Info

Product

Resources

About