10,909 Hits in 4.4 sec

Explainable AI for Natural Adversarial Images [article]

Tomas Folke, ZhaoBin Li, Ravi B. Sojitra, Scott Cheng-Hsin Yang, Patrick Shafto
2021 arXiv   pre-print
Here we evaluate if methods from explainable AI can disrupt this assumption to help participants predict AI classifications for adversarial and standard images.  ...  Adversarial images highlight how vulnerable modern image classifiers are to perturbations outside of their training set.  ...  Here we test whether explanations help people to predict misclassifications of natural adversarial images, which is an essential prerequisite for effective human oversight of AI systems.  ... 
arXiv:2106.09106v1 fatcat:huvnzephw5bv5fo5mq2rja76ri

Overlooked Trustworthiness of Explainability in Medical AI [article]

Jiajin Zhang, Hanqing Chao, Mannudeep K Kalra, Ge Wang, Pingkun Yan
2021 medRxiv   pre-print
AI researchers, practitioners, and authoritative agencies in the medical domain should use caution when explaining AI models because such an explanation could be irrelevant, misleading, and even adversarially  ...  While various methods have been proposed to explain AI models, the trustworthiness of the generated explanation received little examination.  ...  Saliency maps, also commonly referred to as heat maps, are the most commonly used method for AI explainability 6 .  ... 
doi:10.1101/2021.12.23.21268289 fatcat:wg5rbtzmafg7xcmfu5nzuknkte

The effectiveness of feature attribution methods and its correlation with automatic evaluation scores [article]

Giang Nguyen, Daeyoung Kim, Anh Nguyen
2022 arXiv   pre-print
natural or adversarial (i.e., contains adversarial perturbations).  ...  Explaining the decisions of an Artificial Intelligence (AI) model is increasingly critical in many real-world, high-stake applications.  ...  Figure 2 : 3 - 23 Figure 2: 3-NN is consistently among the most effective in improving human-AI team accuracy (%) on natural ImageNet images (a), natural Dog images (b), and adversarial ImageNet images  ... 
arXiv:2105.14944v5 fatcat:iqcul3btbzgepgzdr7drkltrtu

Trustworthy AI: A Computational Perspective [article]

Haochen Liu, Yiqi Wang, Wenqi Fan, Xiaorui Liu, Yaxin Li, Shaili Jain, Yunhao Liu, Anil K. Jain, Jiliang Tang
2021 arXiv   pre-print
In this work, we focus on six of the most crucial dimensions in achieving trustworthy AI: (i) Safety & Robustness, (ii) Non-discrimination & Fairness, (iii) Explainability, (iv) Privacy, (v) Accountability  ...  In this survey, we present a comprehensive survey of trustworthy AI from a computational perspective, to help readers understand the latest technologies for achieving trustworthy AI.  ...  For training in adversarial defense methods, an important issue is the robust overfitting and lack of generalization in both adversarial and natural examples.  ... 
arXiv:2107.06641v3 fatcat:ymqaxvzsoncqrcosj5mxcvgsuy

An Assessment of Robustness for Adversarial Attacks and Physical Distortions on Image Classification using Explainable AI

K. T. Yasas Mahima, Mohamed Ayoob, Guhanathan Poravi
2021 SGAI Conferences  
As a result, research on model interpretability and explainability has been carried out in the domain which is collectively known as Explainable AI.  ...  Using a set of Explainable AI techniques, this study is investigating the deep learning networks' robustness; i.e., the decision-making process in neural networks and important pixel attributes for the  ...  Suresha Perera (Lecturer at the Open University Sri Lanka) for proofreading and guiding the revisions to this paper.  ... 
dblp:conf/sgai/MahimaAP21 fatcat:rdwwpu3eufcujarr4m7uh2a7ty

Securing AI-based Healthcare Systems using Blockchain Technology: A State-of-the-Art Systematic Literature Review and Future Research Directions [article]

Rucha Shinde, Shruti Patil, Ketan Kotecha, Vidyasagar Potdar, Ganeshsree Selvachandran, Ajith Abraham
2022 arXiv   pre-print
We found that 1) Defence techniques for adversarial attacks on AI are available for specific kind of attacks and even adversarial training is AI based technique which in further prone to different attacks  ...  A global solution for all sort of adversarial attacks on AI based healthcare. However, this technique has significant limits and challenges that need to be addressed in future studies.  ...  Adversarial images mislead DNN as a minute perturbation in the input makes DNN vulnerable. The Figure 13 represents the approaches for an adversarial attack on an image.  ... 
arXiv:2206.04793v1 fatcat:v2wrluwugja65btmjct5wlrfm4

Dissociable neural representations of adversarially perturbed images in convolutional neural networks and the human brain [article]

Chi Zhang, Xiaohan Duan, Linyuan Wang, Yongli Li, Bin Yan, Guoen Hu, Ruyuan Zhang, Li Tong
2020 arXiv   pre-print
Here, we leverage adversarial noise (AN) and adversarial interference (AI) images to quantify the consistency between neural representations and perceptual outcomes in the two systems.  ...  In AlexNet, however, the neural representations of adversarial images are inconsistent with network outputs in all intermediate processing layers, providing no neural foundations for perceptual similarity  ...  Informed consent was obtained from all patients for being included in the study.  ... 
arXiv:1812.09431v3 fatcat:k4ps4yljcbexthcul4r36gmahy

Ten AI Stepping Stones for Cybersecurity [article]

Ricardo Morla
2019 arXiv   pre-print
the implications of the adversarial nature of cybersecurity in the learning techniques.  ...  We then discuss the use of AI by attackers on a level playing field including several issues in an AI battlefield, and an AI perspective on the old cat-and-mouse game including how the adversary may assess  ...  In fact, explainability and resilience to adversarial attacks are two of of the challenges identified by UC Berkeley [40] for mission critical AI usage (which includes cybersecurity), together with continuous  ... 
arXiv:1912.06817v1 fatcat:ujeitni5grcopbhl6dwenncjje

Algorithms in future insurance markets

Małgorzata Śmietanka, Adriano Koshiyama, Philip Treleaven
2021 International Journal of Data Science and Big Data Analytics  
., long short-term memory, generative adversarial networks) and leverage important applications (e.g., Natural Language Processing, Adversarial Examples, Deep Fakes, etc.).  ...  These technologies are important since they underpin the automation of the insurance markets and risk analysis, and provide the context for the algorithms, such as AI machine learning and computational  ...  fairness), and (b) those related to the dynamic nature of AI.  ... 
doi:10.51483/ijdsbda.1.1.2021.1-19 fatcat:gty5qdugnbhm3mophojqyxmkja

Developing Future Human-Centered Smart Cities: Critical Analysis of Smart City Security, Interpretability, and Ethical Challenges [article]

Kashif Ahmad, Majdi Maabreh, Mohamed Ghaly, Khalil Khan, Junaid Qadir, Ala Al-Fuqaha
2021 arXiv   pre-print
Globally there are calls for technology to be made more humane and human-centered.  ...  We believe such rigorous analysis will provide a baseline for future research in the domain.  ...  " in nature.  ... 
arXiv:2012.09110v4 fatcat:yxh5tvpehbgldcblweoovbvdsq

Robustness of different loss functions and their impact on networks learning capability [article]

Vishal Rajput
2021 arXiv   pre-print
For our case, we will use two sets of loss functions, generalized loss functions like Binary cross-entropy or BCE and specialized loss functions like Dice loss or focal loss.  ...  In order to establish the difference between generalized loss and specialized losses, we will train several models using the above-mentioned losses and then compare their robustness on adversarial examples  ...  Explainability in AI leads to Responsible Artificial Intelligence, namely, a methodology for the large-scale implementation of AI solutions in real-life scenarios with fairness, explainability and accountability  ... 
arXiv:2110.08322v2 fatcat:qzruirgisrgjbgnwfen64fep7m

When Explainability Meets Adversarial Learning: Detecting Adversarial Examples using SHAP Signatures

Gil Fidel, Ron Bitton, Asaf Shabtai
2019 Zenodo  
In this research, we present a novel detection method that uses Shapley Additive Explanations (SHAP) values computed for the internal layers of a DNN classifier to discriminate between normal and adversarial  ...  and adversarial inputs.  ...  We try to leverage this hypothesized property of adversarial examples by utilizing explainable AI methods (XAI) for interpreting model predictions.  ... 
doi:10.5281/zenodo.3941781 fatcat:xea2tuzzmngtrgjvorhwqky5lm

Towards Resilient Artificial Intelligence: Survey and Research Issues [article]

Oliver Eigner, Sebastian Eresheim, Peter Kieseberg, Lukas Daniel Klausner, Martin Pirker, Torsten Priebe, Simon Tjoa, Fiammetta Marulli, Francesco Mercaldo
2021 arXiv   pre-print
Considering the particular nature of AI, and machine learning (ML) in particular, this paper provides an overview of the emerging field of resilient AI and presents research issues the authors identify  ...  Their resilience against attacks and other environmental influences needs to be ensured just like for other IT assets.  ...  Due to the explainability problem, even proper annotation of the changes made does not fully explain the impact on AI-based decisions; thus the question of responsibility for errors and negative effects  ... 
arXiv:2109.08904v1 fatcat:vadq2vohljhxpbcokklir4buee

A Review on Explainability in Multimodal Deep Neural Nets

Gargi Joshi, Rahee Walambe, Ketan Kotecha
2021 IEEE Access  
This has given rise to the quest for model interpretability and explainability, more so in the complex tasks involving multimodal AI methods.  ...  INDEX TERMS deep multimodal learning, explainable AI, interpretability, survey, trends, vision and language research, XAI.  ...  AI (XAI) approaches for the Natural Language Processing domain.  ... 
doi:10.1109/access.2021.3070212 fatcat:5wtxr4nf7rbshk5zx7lzbtcram

An Adversarial Approach for Explaining the Predictions of Deep Neural Networks [article]

Arash Rahnama, Andrew Tseng
2020 arXiv   pre-print
In this work, we present a novel algorithm for explaining the predictions of a DNN using adversarial machine learning.  ...  We present our detailed analysis that demonstrates how the behavior of an adversarial attack, given a DNN and a task, stays consistent for any input test data point proving the generality of our approach  ...  Images from CIFAR10 dataset are easily explained due to the nature of the dataset with most objects in the images being located in the middle of the image and the lack of noisy background in most images  ... 
arXiv:2005.10284v4 fatcat:7p3pi6qn75eoxlpdgflumam4hi
« Previous Showing results 1 — 15 out of 10,909 results