AIPA: An Adversarial Imperceptible Patch Attack on Medical Datasets and its Interpretability

Article Automation & Control Systems

Securing Facial Bioinformation by Eliminating Adversarial Perturbations

Feng Ding et al.

Summary: Falsified faces generated by DeepFake pose severe threats to our community, especially to smart systems relying on bioinformation authentication. Despite promising results on DeepFake forensics, new security challenges arise from antiforensics attacks. To protect biometric data, particularly facial information, a countermeasure against DeepFake antiforensics attacks is proposed.

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS (2023)

Add to Collection

Article Computer Science, Artificial Intelligence

A large multiclass dataset of CT scans for COVID-19 identification

Eduardo Soares et al.

Summary: The SARS-CoV-2 infection, causing the COVID-19 disease, has rapidly spread worldwide since the start of 2020. The World Health Organization (WHO) declared a global health emergency and a pandemic following the outbreak in Wuhan, China on January 30, 2020. This paper presents a publicly available multiclass CT scan dataset for identifying SARS-CoV-2 infection. The dataset consists of 4173 CT scans from 210 different patients, including 2168 scans from 80 confirmed SARS-CoV-2 infected patients. The aim of this dataset is to facilitate research and development of artificial intelligence methods for CT scan analysis in identifying SARS-CoV-2 and other diseases. The eXplainable Deep Learning approach (xDNN) was used as the baseline method for this dataset, providing transparency in network decision-making.

EVOLVING SYSTEMS (2023)

Add to Collection

Review Dentistry, Oral Surgery & Medicine

Deep learning models in medical image analysis

Masayuki Tsuneki

Summary: Deep learning is a powerful technology for medical image analysis that offers fast and robust object detection, segmentation, tracking, and classification of pathophysiological anatomical structures. However, the lack of sufficient medical images for training sets poses a major limitation for deep learning in medical image analysis.

JOURNAL OF ORAL BIOSCIENCES (2022)

Add to Collection

Article Computer Science, Information Systems

Mitigating adversarial evasion attacks by deep active learning for medical image classification

Usman Ahmed et al.

Summary: Collaboration among institutes in the Internet of Medical Things (IoMT) can assist in complex medical and clinical analysis of diseases. This research proposes institutional data collaboration and an adversarial evasion method to enhance the availability of diverse training data and protect sensitive information. The model successfully evades attacks and achieves a high accuracy rating of 95%.

MULTIMEDIA TOOLS AND APPLICATIONS (2022)

Add to Collection

Review Computer Science, Artificial Intelligence

The transformational role of GPU computing and deep learning in drug discovery

Mohit Pandey et al.

Summary: This review summarizes the significant role GPUs have played in advancing deep learning methods in drug discovery, including their applications in accelerating molecular docking, evaluating off-target effects, and predicting pharmacological properties.

NATURE MACHINE INTELLIGENCE (2022)

Add to Collection

Review Physics, Multidisciplinary

Explainable AI: A Review of Machine Learning Interpretability Methods

Pantelis Linardatos et al.

Summary: Recent advances in artificial intelligence have led to widespread industrial adoption, with machine learning systems demonstrating superhuman performance. However, the complexity of these systems has made them difficult to explain, hindering their application in sensitive domains. Therefore, there is a renewed interest in the field of explainable artificial intelligence.

ENTROPY (2021)

Add to Collection

Article Computer Science, Information Systems

Project Gradient Descent Adversarial Attack against Multisource Remote Sensing Image Scene Classification

Yan Jiang et al.

Summary: Deep learning and remote sensing technology have rapidly developed, with deep convolutional neural networks achieving state-of-the-art performance in scene classification. However, adversarial attacks pose a potential threat to classification accuracy, and a projected gradient descent method is introduced to generate adversarial remote sensing images, significantly reducing classification accuracy. The quality of the generated adversarial images is imperceptible, demonstrating a stronger attack ability in remote sensing scene classification.

SECURITY AND COMMUNICATION NETWORKS (2021)

Add to Collection

Article Computer Science, Information Systems