Article
Computer Science, Artificial Intelligence
El Houssaine Hssayni, Nour-Eddine Joudar, Mohamed Ettaouil
Summary: This article introduces a neural network regularization method called DropConnect and its generalized version, A-DropConnect. By estimating the DropConnect hyperparameter using the gap generalization and the Rademacher complexity, the A-DropConnect technique is proposed and demonstrated to be effective through experiments.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Pieter Van Molle, Tim Verbelen, Bert Vankeirsbilck, Jonas De Vylder, Bart Diricx, Tom Kimpe, Pieter Simoens, Bart Dhoedt
Summary: The paper highlights the limitations of conventional neural networks in capturing uncertainty and introduces Bayesian techniques such as Monte Carlo dropout. The authors propose a novel method based on the overlap of output distributions of different classes to better approximate inter-class output confusion. They demonstrate the advantages of their approach using benchmark datasets and skin lesion classification.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Jakob Gawlikowski, Cedrique Rovile Njieutcheu Tassi, Mohsin Ali, Jongseok Lee, Matthias Humt, Jianxiang Feng, Anna Kruspe, Rudolph Triebel, Peter Jung, Ribana Roscher, Muhammad Shahzad, Wen Yang, Richard Bamler, Xiao Xiang Zhu
Summary: Neural networks have become integral to various real world applications, but their predictions lack certainty estimates and suffer from calibration issues. Researchers have been focusing on understanding and quantifying uncertainty in neural network predictions, resulting in the identification of different types and sources of uncertainty and the proposal of various approaches to measure and quantify it. This work provides a comprehensive overview of uncertainty estimation in neural networks, reviews recent advances, discusses challenges, and identifies research opportunities.
ARTIFICIAL INTELLIGENCE REVIEW
(2023)
Article
Engineering, Industrial
Ramin Moradi, Sergio Cofre-Martel, Enrique Lopez Droguett, Mohammad Modarres, Katrina M. Groth
Summary: This paper presents a novel mathematical architecture for risk and reliability analysis of complex engineering systems, addressing both the complexity of operational data and system complexity using Bayesian networks and Bayesian deep learning models, providing system-level insights.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2022)
Article
Computer Science, Artificial Intelligence
Minjung Lee, Jinsoo Bae, Seoung Bum Kim
Summary: Data-driven soft sensors using deep learning models have shown superior predictive performance, but may face trustworthiness issues when dealing with unexpected situations or noisy input data. By introducing uncertainty-aware soft sensors based on Bayesian recurrent neural networks, the reliability of predictive uncertainty can be increased, allowing for interval prediction without compromising the predictive performance of the soft sensor.
ADVANCED ENGINEERING INFORMATICS
(2021)
Article
Engineering, Multidisciplinary
Gabriel D. Maher, Casey M. Fleeter, Daniele E. Schiavazzi, Alison L. Marsden
Summary: A novel approach utilizing convolutional neural networks is proposed to generate samples from patient-specific cardiovascular models based on clinically acquired image volumes. The method focuses on learning geometric uncertainty directly from training data and demonstrates its impact on hemodynamics for different patient-specific anatomies.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Artificial Intelligence
Pier Luigi Conti, Daniela Marella, Paola Vicard, Vincenzina Vitale
Summary: The use of Bayesian networks in statistical matching allows for introducing additional sample information on the dependence structure between variables and simplifying parameter estimation, leading to improved matching quality in a multivariate context.
INTERNATIONAL JOURNAL OF APPROXIMATE REASONING
(2021)
Article
Mechanics
Nils Wandel, Michael Weinmann, Reinhard Klein
Summary: This work introduces significant extensions to a deep learning framework for fluid simulations, enabling fast and accurate 3D simulations while conditioning the neural fluid model on additional information about viscosity and density. The method allows for simulating laminar and turbulent flows without prior fluid simulation data, showing improvements in accuracy, speed, and generalization capabilities compared to current 3D NN-based fluid models.
Article
Astronomy & Astrophysics
B. Acharya, S. Bacca
Summary: In this study, the energy-dependent cross section of the np ? d gamma process is calculated in chiral effective field theory, and state-of-the-art tools are used to quantify the theory uncertainty. The Gaussian process error model is found to describe the observed convergence well, and Bayesian credible intervals for the truncation error are presented.
Article
Computer Science, Artificial Intelligence
Hampus Linander, Oleksandr Balabanov, Henry Yang, Bernhard Mehlig
Summary: Bayesian inference can quantify uncertainty in neural network predictions using posterior distributions, and we show how prediction accuracy is related to epistemic and aleatoric uncertainties. We also introduce a novel acquisition function that outperforms common methods.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2023)
Article
Construction & Building Technology
Lorin Werthen-Brabants, Tom Dhaene, Dirk Deschrijver
Summary: This paper investigates the uncertainty issue in Non-Intrusive Load Monitoring (NILM). It shows that uncertainty information about model misclassifications can be obtained using Bayesian Neural Networks, which is valuable for end-users. The experiment demonstrates an improvement in model generalization performance by using Stochastic Gradient Hamiltonian Monte Carlo.
ENERGY AND BUILDINGS
(2022)
Article
Public, Environmental & Occupational Health
Laura Urso, Mouhamadou Moustapha Sy, Marc-Andre Gonze, Philipp Hartmann, Martin Steiner
Summary: This study focuses on assessing conceptual model uncertainty, subtracting parameter uncertainty obtained through Bayesian inference analysis from the total uncertainty of the model output for two process-based models describing the interception of wet deposited pollutants. Quantitative evidence suggests that conceptual model uncertainty can contribute as much as, if not more than, parameter uncertainty to the total uncertainty budget of the models.
Article
Nuclear Science & Technology
Lesego E. Moloko, Pavel M. Bokov, Xu Wu, Kostadin N. Ivanov
Summary: This study uses Deep Neural Networks (DNNs) to predict assembly axial neutron flux profiles in the SAFARI-1 research reactor and quantifies the uncertainties in DNN predictions. Uncertainty Quantification is done using Monte Carlo Dropout (MCD) and Bayesian Neural Networks solved by Variational Inference (BNN VI). The results show that regular DNNs, DNNs with MCD, and BNN VI all have good prediction and generalization capabilities, and the uncertainty bands produced by MCD and BNN VI accurately envelope the measurement data points.
ANNALS OF NUCLEAR ENERGY
(2023)
Article
Computer Science, Artificial Intelligence
Yufeng Xia, Jun Zhang, Tingsong Jiang, Zhiqiang Gong, Wen Yao, Ling Feng
Summary: HatchEnsemble improves the efficiency and practicality of Deep Ensemble by using HatchNets to inherit the knowledge learned by a single model SeedNet. Experiments show competitive prediction performance and better-calibrated uncertainty quantification compared to baselines.
COMPLEX & INTELLIGENT SYSTEMS
(2021)
Article
Materials Science, Multidisciplinary
Minglang Yin, Zongren Zou, Enrui Zhang, Cristina Cavinato, Jay D. Humphrey, George Em Karniadakis
Summary: This research proposes a novel approach that combines generative deep learning with Bayesian inference to efficiently infer families of constitutive relationships in data-sparse regimes, which could deepen our understanding of the biomechanical properties of the human vasculature.
JOURNAL OF THE MECHANICS AND PHYSICS OF SOLIDS
(2023)
Meeting Abstract
Infectious Diseases
A. T. Nguyen, N. Le, H. Nguyen, T. Tran, C. Anscombe, C. -Y. Lau, D. Limmathurotsakul, C. Nguyen, R. van Doorn, X. Deng, M. Rahman, E. Delwart, G. Thwaites, T. Le
INTERNATIONAL JOURNAL OF INFECTIOUS DISEASES
(2019)
Article
Medicine, General & Internal
Meghana Padmanabhan, Pengyu Yuan, Govind Chada, Hien Van Nguyen
JOURNAL OF CLINICAL MEDICINE
(2019)
Article
Medicine, General & Internal
Aryan Mobiny, Aditi Singh, Hien Van Nguyen
JOURNAL OF CLINICAL MEDICINE
(2019)
Article
Medicine, General & Internal
Pengyu Yuan, Ali Rezvan, Xiaoyang Li, Navin Varadarajan, Hien Van Nguyen
JOURNAL OF CLINICAL MEDICINE
(2019)
Article
Computer Science, Interdisciplinary Applications
Aryan Mobiny, Hengyang Lu, Hien V. Nguyen, Badrinath Roysam, Navin Varadarajan
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2020)
Review
Urology & Nephrology
Jan U. Becker, David Mayerich, Meghana Padmanabhan, Jonathan Barratt, Angela Ernst, Peter Boor, Pietro A. Cicalese, Chandra Mohan, Hien V. Nguyen, Badrinath Roysam
KIDNEY INTERNATIONAL
(2020)
Article
Engineering, Multidisciplinary
Furui Wang, Aryan Mobiny, Hien Van Nguyen, Gangbing Song
Summary: A novel percussion method was developed to detect spatial bolt looseness, achieving accurate detection of multi-bolt looseness for the first time with the use of a memory-augmented neural network. This approach effectively avoids inefficient relearning, improves detection accuracy, and preliminarily explores the potential of implementing automation applications in real industries.
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2021)
Article
Computer Science, Information Systems
Pietro Antonio Cicalese, Aryan Mobiny, Zahed Shahmoradi, Xiongfeng Yi, Chandra Mohan, Hien Van Nguyen
Summary: The diagnosis of Lupus Nephritis (LN) based on kidney biopsy suffers from low inter-observer agreement, and Computer Aided Diagnosis (CAD) systems have not yet been effectively utilized for accurate classification. The proposed Uncertainty-Guided Bayesian Classification (UGBC) scheme aims to improve classification accuracy, utilizing deep neural networks and achieving promising results.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Aryan Mobiny, Pengyu Yuan, Pietro A. Cicalese, Supratik K. Moulik, Naveen Garg, Carol C. Wu, Kelvin Wong, Stephen T. Wong, Tian Cheng He, Hien V. Nguyen
Summary: The proposed memory-augmented capsule network enables rapid adaptation of CAD models to new domains with few annotated samples, achieving significant performance improvements while reducing the use of labeled training data. Evaluation with a large-scale public dataset demonstrates the robustness of the method under severe conditions compared to state-of-the-art approaches.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Samira Zare, Hien Van Nguyen
Summary: While deep networks have achieved excellent performance in medical image analysis, they are influenced by biases caused by confounding variables. Traditional statistical methods are not compatible with deep networks. To address this issue, we propose a novel learning framework called ReConfirm, based on the IRM theory, to eliminate biases caused by confounding variables and enhance the robustness of deep networks. We evaluate our approach on NIH chest X-ray classification tasks with sex and age as confounding variables.
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION, MICCAI 2022, PT VIII
(2022)
Proceedings Paper
Acoustics
Pietro Antonio Cicalese, Syed Asad Rizvi, Victor Wang, Sai Patibandla, Pengyu Yuan, Samira Zare, Katharina Moos, Ibrahim Batal, Marian Clahsen-van Groningen, Candice Roufosse, Jan Becker, Chandra Mohan, Hien Van Nguyen
Summary: Computer Aided Diagnosis (CAD) systems using Deep neural networks (DNNs) show promise in improving diagnostic accuracy in renal pathology. The introduction of MorphSet architecture achieved 98.9% case level accuracy, surpassing traditional consensus labeling methods.
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION - MICCAI 2021, PT VIII
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Xinyue Zhang, Jiahao Ding, Maoqiang Wu, Stephen T. C. Wong, Hien Van Nguyen, Miao Pan
Summary: Deep learning has the potential to revolutionize healthcare and medicine, but also poses risks to patient information privacy. This paper proposes a novel privacy-preserving mechanism by adding decaying Gaussian noise to gradients, which achieves significantly higher classification accuracy compared to existing methods.
2021 IEEE WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV 2021)
(2021)
Proceedings Paper
Acoustics
Ilker Gurcan, Hien Van Nguyen
2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2019)
Article
Chemistry, Analytical
Sebastian Berisha, Mahsa Lotfollahi, Jahandar Jahanipour, Ilker Gurcan, Michael Walsh, Rohit Bhargava, Hien Van Nguyen, David Mayerich