Article
Computer Science, Information Systems
Weiping Ding, Mohamed Abdel-Basset, Hossam Hawash, Ahmed M. Ali
Summary: The continuous advancement of Artificial Intelligence (AI) has revolutionized decision-making in various domains, but the lack of transparency and explainability in AI algorithms poses ethical challenges. Explainable Artificial Intelligence (XAI) aims to generate human-comprehensible explanations to reveal the internal workings of AI decisions. This study provides a taxonomy and evaluation of XAI research, discusses the advantages, limitations, and evaluation metrics of explanation generation techniques, and identifies future research directions and challenges.
INFORMATION SCIENCES
(2022)
Article
Physics, Multidisciplinary
Kaleel Mahmood, Deniz Gurevin, Marten van Dijk, Phuoung Ha Nguyen
Summary: The study expands upon the analyses of recent defenses and evaluates them against adaptive black-box adversaries. The results show that most recent defenses only provide slight improvements in security, highlighting the need for thorough white-box and black-box analyses to be considered secure.
Article
Computer Science, Information Systems
Ibomoiye Domor Mienye, Yanxia Sun
Summary: Ensemble learning techniques have achieved state-of-the-art performance by combining predictions from multiple base models, with a focus on widely used algorithms such as random forest, AdaBoost, gradient boosting, XGBoost, LightGBM, and CatBoost. This overview aims to provide concise coverage of their mathematical and algorithmic representations, lacking in existing literature, for the benefit of machine learning researchers and practitioners.
Article
Computer Science, Artificial Intelligence
Manish Narwaria
Summary: Machine learning and deep learning are widely used in vision applications, but the issue of explainability has limitations and may not effectively uncover black box models. To improve explainability, more rigorous principles in related areas should be relied upon.
IMAGE AND VISION COMPUTING
(2022)
Review
Medical Informatics
Wei Yan Ng, Tien-En Tan, Prasanth V. H. Movva, Andrew Hao Sen Fang, Khung-Keong Yeo, Dean Ho, Fuji Shyy San Foo, Zhe Xiao, Kai Sun, Tien Yin Wong, Alex Tiong-Heng Sia, Daniel Shu Wei Ting
Summary: Blockchain technology has the potential for various applications in healthcare, both related to COVID-19 and unrelated. However, current research mostly focuses on technical aspects with few real-world clinical applications, underscoring the need to translate foundational blockchain technology into clinical use.
LANCET DIGITAL HEALTH
(2021)
Article
Computer Science, Information Systems
Miles Q. Li, Benjamin C. M. Fung, Philippe Charland
Summary: Recent research shows that machine learning models are susceptible to attacks from slightly perturbed adversarial samples. These attacks can be carried out in white-box or black-box scenarios. Existing defense methods are static and cannot adapt to adversarial attacks. This paper introduces a novel dynamic defense method called DyAdvDefender, which effectively utilizes previous experience to defend against black-box attacks.
INFORMATION SCIENCES
(2022)
Article
Physics, Multidisciplinary
Hong Zhao
Summary: The study shows that by using a self-evolution learning machine with a training strategy that gradually decreases the cost function, it is possible to infer the dynamic behavior of a system and gradually reveal its dynamic properties during training.
SCIENCE CHINA-PHYSICS MECHANICS & ASTRONOMY
(2021)
Article
Green & Sustainable Science & Technology
Indira Adilkhanova, Jack Ngarambe, Geun Young Yun
Summary: The whitebox models are popular due to their transparent working process and high prediction accuracy, but some simulation tools are too complex and require a high level of expertise to operate, potentially leading to inaccuracies in the outcomes. On the other hand, black-box models, despite their opaque working process, are easier to use and require less computation time. The fusion of these two methods is a novel approach that may benefit both UHI prediction and mitigation at the design and operation stages.
RENEWABLE & SUSTAINABLE ENERGY REVIEWS
(2022)
Article
Computer Science, Theory & Methods
Luca Demetrio, Battista Biggio, Giovanni Lagorio, Fabio Roli, Alessandro Armando
Summary: This paper introduces a novel family of black-box attacks that are both query-efficient and functionality-preserving, by injecting benign content into malicious files. Empirical investigation shows that these black-box attacks can bypass popular static Windows malware detectors with few queries and small payloads, even when they only return predicted labels. Surprisingly, the attacks can also evade more than 12 commercial antivirus engines on average.
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
(2021)
Article
Allergy
Danielle R. Stevens, Neil Perkins, Zhen Chen, Rajesh Kumar, William Grobman, Akila Subramaniam, Joseph Biggio, Katherine L. Grantz, Seth Sherman, Matthew Rohn, Pauline Mendola
Summary: This study aimed to identify and characterize trajectories of gestational asthma control in a US-based prospective pregnancy cohort. Two trajectories were identified: one with worsened asthma control and one with stable asthma control.
JOURNAL OF ALLERGY AND CLINICAL IMMUNOLOGY-IN PRACTICE
(2022)
Review
Computer Science, Artificial Intelligence
Piero Fraternali, Federico Milani, Rocio Nahime Torres, Niccolo Zangrando
Summary: The application of DNNs to various tasks requires methods to deal with their complex and opaque nature. Evaluating performance beyond standard metrics, understanding model behavior, and diagnosing prediction errors can be achieved through model interpretation and black-box error diagnosis techniques. Both approaches provide insights for improving the architecture and training process.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Kaleel Mahmood, Phuong Ha Nguyen, Lam M. Nguyen, Thanh Nguyen, Marten Van Dijk
Summary: Most existing adversarial machine learning defenses are designed to mitigate static, white-box attacks, but their robustness against adaptive black-box attacks is still unknown. This paper focuses on the black-box threat model and makes two main contributions: first, it proposes an enhanced adaptive black-box attack that is significantly more effective than previous approaches; second, it tests 10 recent defenses and introduces a new black-box defense called barrier zones, which demonstrates significant improvements in security.
Article
Computer Science, Information Systems
Michele Polese, Rittwik Jana, Velin Kounev, Ke Zhang, Supratim Deb, Michele Zorzi
Summary: This paper explores the potential of using edge cloud deployments in 5G cellular networks to enable intelligent data and machine learning applications. It proposes an edge-controller-based network architecture, evaluates its performance with real data, and demonstrates the application of machine learning algorithms in predicting user numbers.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2021)
Article
Computer Science, Information Systems
Kaleel Mahmood, Rigel Mahmood, Ethan Rathbun, Marten van Dijk
Summary: This paper provides a systematic survey and categorization of recent advances in adversarial machine learning black-box attacks. It summarizes 20 black-box attacks and introduces a new mathematical framework for evaluating attack results.
Article
Computer Science, Artificial Intelligence
Xiaohang Zhang, Yuan Wang, Zhengren Li
Summary: This study introduces a method to gain insight into black boxes of supervised learning models by visualizing the impacts of input features on prediction results. The method distinguishes typical impact patterns corresponding to different groups of observations, and reveals feature relationships embedded in prediction models.
APPLIED INTELLIGENCE
(2021)
Article
Economics
Sean P. Gavan, Ian N. Bruce, Katherine Payne
Summary: This study aimed to estimate the value of health gain for patients with a multisystem disease using specific composite response endpoints. Data from a national register were used to compare health utility values and quality-adjusted life-years for responders and nonresponders. The study found that bespoke composite response endpoints can be used to assess treatment response for multisystem diseases and provide valuable evidence for health technology assessment, decision making, and economic evaluation.
Article
Communication
Huw Roberts, Josh Cowls, Emmie Hine, Jessica Morley, Vincent Wang, Mariarosaria Taddeo, Luciano Floridi
Summary: This article compares the artificial intelligence strategies of China and the European Union, focusing on their high-level aims, approaches to promoting AI development and use, and the intended beneficiaries of these policies. By examining the similarities and differences, the article suggests areas where both the EU and China can learn from each other and improve their AI governance to achieve more ethical outcomes. Policy recommendations are provided for European and Chinese policymakers.
INFORMATION SOCIETY
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
Marta Tibiletti, James A. Eaden, Josephine H. Naish, Paul J. C. Hughes, John C. Waterton, Matthew J. Heaton, Nazia Chaudhuri, Sarah Skeoch, Ian N. Bruce, Stephen Bianchi, Jim M. Wild, Geoff J. M. Parker
Summary: This study aimed to compare imaging biomarkers from hyperpolarised 129Xe ventilation MRI and dynamic oxygen-enhanced MRI (OE-MRI) with standard pulmonary function tests (PFT) in interstitial lung disease (ILD) patients. The results showed that the ventilation biomarkers were not able to differentiate between ILD subtypes and had limited correlation with PFT changes.
MAGNETIC RESONANCE IMAGING
(2023)
Article
Computer Science, Artificial Intelligence
Jakob Mokander, Margi Sheth, David S. Watson, Luciano Floridi
Summary: There is a gap between principles and practices in AI ethics due to the lack of a well-defined material scope. This article reviews and compares previous attempts to classify AI systems using three mental models: the Switch, the Ladder, and the Matrix. Each model has its own strengths and weaknesses. By conceptualizing different ways of classifying AI systems, organizations can demarcate the material scope of their AI governance frameworks.
MINDS AND MACHINES
(2023)
Article
Computer Science, Artificial Intelligence
Emmie Hine, Luciano Floridi
Summary: The US is promoting a vision of a Good AI Society through its AI Bill of Rights, emphasizing community-oriented equity unique amongst its peers. However, there is a risk of rights violations and the non-binding nature of the bill may lead the private sector to ignore it.
MINDS AND MACHINES
(2023)
Letter
Health Care Sciences & Services
Ian N. Bruce, Sarowar Golam, Jason Steenkamp, Pearl Wang, Evelyn Worthington, Barnabas Desta, Konstantina Psachoulia, Wilma Erhardt, Raj Tummala
JOURNAL OF COMPARATIVE EFFECTIVENESS RESEARCH
(2023)
Article
Rheumatology
Ian N. Bruce, Ronald F. van Vollenhoven, Konstantina Psachoulia, Catharina Lindholm, Emmanuelle Maho, Raj Tummala
Summary: This study evaluated the time course of clinical response following anifrolumab treatment in SLE patients and found that anifrolumab treatment improved overall disease activity and skin responses from Week 8, leading to glucocorticoid reductions from Week 20. These findings provide insights for physicians and patients on the potential clinical response timeline for anifrolumab treatment.
LUPUS SCIENCE & MEDICINE
(2023)
Article
Statistics & Probability
Kristin Blesch, David S. Watson, Marvin N. Wright
Summary: Despite the popularity of feature importance (FI) measures in interpretable machine learning, the statistical adequacy of these methods is rarely discussed. Our work draws attention to the distinction between marginal and conditional FI measures and their implications. We propose a workflow that combines the conditional predictive impact (CPI) framework with sequential knockoff sampling to provide statistically adequate measurement of conditional FI for mixed data.
ASTA-ADVANCES IN STATISTICAL ANALYSIS
(2023)
Article
Medicine, General & Internal
Michelle Petri, Ian N. Bruce, Thomas Doerner, Yoshiya Tanaka, Eric F. Morand, Kenneth C. Kalunian, Mario H. Cardiel, Maria E. Silk, Christina L. Dickson, Gabriella Meszaros, Lu Zhang, Bochao Jia, Youna Zhao, Conor J. McVeigh, Marta Mosca
Summary: Baricitinib, an oral inhibitor of Janus kinase 1 and 2, was evaluated as a treatment for patients with systemic lupus erythematosus (SLE) in a 52-week phase 3 study. The study found that baricitinib did not significantly improve SLE disease activity compared with placebo, and there were no new safety concerns.
Article
Rheumatology
M. L. Barraclough, J. P. Diaz-Martinez, A. Knight, K. Bingham, J. Su, M. Kakvan, C. Munoz Grajales, M. C. Tartaglia, L. Ruttan, J. Wither, M. Y. Choi, D. Bonilla, N. Anderson, S. Appenzeller, B. Parker, P. Katz, D. Beaton, R. Green, I. N. Bruce, Z. Touma
Summary: During the COVID-19 pandemic, our longitudinal study on cognitive impairment in systemic lupus erythematosus (SLE) had to adapt and switch from in-person to virtual administration. We found that the method of administration had an impact on cognitive performance and classification. Approximately 42% of the tests in the battery encountered issues when moved from in-person to virtual administration, highlighting the need for caution when comparing results.
Correction
Computer Science, Artificial Intelligence
Jakob Mokander, Margi Sheth, David S. Watson, Luciano Floridi
MINDS AND MACHINES
(2023)
Article
Medicine, General & Internal
Callum Andrew Shields, Mark Sladen, Azita Rajai, Hannah Guest, Iain Bruce, Karolina Kluk, Jaya Nichani
Summary: This study aims to explore the use of existing questionnaire-based outcome instruments to evaluate listening effort (LE) and its associated consequences in children and young people with and without hearing loss. Participants will complete online questionnaires and a hearing test, and the data will be analyzed to identify correlations between LE and other outcomes.
Article
Computer Science, Artificial Intelligence
Claudio Novelli, Mariarosaria Taddeo, Luciano Floridi
Summary: Accountability is essential in the governance of AI, but its definition is often unclear due to the complex nature of AI systems. In this article, we define accountability in terms of answerability and propose a framework consisting of three conditions (authority recognition, interrogation, and limitation of power) and seven features (context, range, agent, forum, standards, process, and implications). We analyze this framework through four accountability goals (compliance, report, oversight, and enforcement) and argue that these goals are complementary and prioritized based on the use of accountability and the objectives of AI governance.
Review
Cell Biology
Sandra Ng, Sara Masarone, David Watson, Michael R. R. Barnes
Summary: The prospects for discovering reliable and reproducible biomarkers have improved greatly with the development of sensitive omics platforms. However, the challenge now lies in the analytical domain, as standard statistical methods struggle to distinguish signal from noise in complex biological systems. Machine learning and AI methods offer potential solutions, but overfitting remains a concern. The rise of explainable AI provides an opportunity to improve the discovery process by explaining predictions and exploring mechanisms before validation studies.
CELL AND TISSUE RESEARCH
(2023)
Article
Rheumatology
Mia Rodziewicz, Sarah Dyball, Mark Lunt, Stephen McDonald, Emily Sutton, Ben Parker, Ian N. Bruce
Summary: This study investigated the frequency and risk factors for serious infections in patients with moderate-to-severe systemic lupus erythematosus (SLE) treated with rituximab, belimumab, and standard therapies. The study found that rituximab, belimumab, and standard therapy had similar risks for serious infections. Key risk factors for serious infections included multimorbidity, hypogammaglobulinemia, and increased glucocorticoid doses. These findings are important for guiding the treatment and management of SLE patients.
LANCET RHEUMATOLOGY
(2023)