Article
Automation & Control Systems
Nurul Afiqah Burhanuddin, Mohd Bakri Adam, Kamarulzaman Ibrahim
Summary: The paper introduces a constrained Dirichlet process mixture model with labels as side information, which can give clusters with similar labels higher preference and handle multiple side information. Empirical results show that the proposed method consistently improves clustering performance with more labeled data, and rarely performs worse than its unsupervised counterpart even with noisy labels.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Engineering, Electrical & Electronic
Xiaoyin Nie, Gang Xie
Summary: This paper introduces a novel deep learning framework, CRRCNN, for fault diagnosis of wind turbines, which includes clean revision to address the potential corruption in health condition dataset. Experimental results demonstrate the superiority of the proposed framework.
IEEE SENSORS JOURNAL
(2021)
Article
Education, Scientific Disciplines
Ji-Yoon Kim, Sung-Bae Cho
Summary: This paper proposes an ensemble classifier composed of diverse convolutional neural networks (CNNs) to predict repayment in social lending. The approach utilizes pseudo-labels and an uncertainty handling scheme to enhance the model's performance. Experimental results show that this diverse ensemble method outperforms conventional methods.
Article
Computer Science, Artificial Intelligence
Xiaoshuan Shi, Zhenhua Guo, Kang Li, Yun Liang, Xiaofeng Zhu
Summary: Noisy labels can significantly degrade the performance of convolutional neural networks (CNNs). This paper proposes a novel self-paced resistance framework to resist corrupted labels, using the memorization effect of CNNs and a resistance loss to update the model parameters. Extensive experiments demonstrate the superior performance of this framework on noisy-label data.
PATTERN RECOGNITION
(2023)
Article
Multidisciplinary Sciences
Fan Zhou, Xiaozhe Meng, Yuxin Feng, Zhuo Su
Summary: This paper proposes a semi-supervised neural process dehazing network that utilizes asymmetric pseudo labels to address the distribution shift between real and synthetic data, leading to improved dehazing results.
Article
Computer Science, Artificial Intelligence
Yayong Li, Jie Yin, Ling Chen
Summary: In this paper, a novel informative pseudo-labeling framework (InfoGNN) is proposed to effectively learn GNNs with very few labels. The key idea is to pseudo-label the most informative nodes via mutual information maximization, and a class-balanced regularization is designed to mitigate label noise and class-imbalance problem. Extensive experiments validate that the proposed approach significantly outperforms state-of-the-art baselines and competitive self-supervised methods on real-world graph datasets.
DATA MINING AND KNOWLEDGE DISCOVERY
(2023)
Article
Computer Science, Artificial Intelligence
Ehsan Adeli, Luning Sun, Jianxun Wang, Alexandros A. Taflanidis
Summary: In this research, artificial neural network models are used to emulate storm surge based on storm history. The developed convolutional recurrent neural network model outperforms the Gaussian process implementation for storm surge predictions.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Chunxu Zhang, Ximing Li, Hongbin Pei, Zijian Zhang, Bing Liu, Bo Yang
Summary: In this study, a label-enhanced network architecture called LAENNet is proposed to address the robustness issue of GCNs in noisy and sparse graph data scenarios. Experimental results demonstrate the superiority of LAENNet over existing baseline models.
Article
Computer Science, Artificial Intelligence
Nassima Dif, Mohammed Oualid Attaoui, Zakaria Elberrichi, Mustapha Lebbah, Hanene Azzag
Summary: This study introduces a new strategy combining unsupervised learning (clustering) and transfer learning. Clustering methods are used to generate synthetic labels for the source dataset, which is then utilized in transfer learning to other histopathological datasets. The efficiency of MOC-Stream in comparison to K-means is demonstrated, showing that the synthetic histopathological dataset generated by this clustering algorithm outperformed the original labeled dataset and imageNet models in transfer learning.
APPLIED INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Zhanglei Shi, Hao Wang, Chi-Sing Leung
Summary: This paper introduces a feature extraction method based on the CCL algorithm. By using a strategy of fixing connection weights and updating cluster centers, and simplifying calculations with the SCCL algorithm, experiments have shown that this method outperforms other approaches on multiple datasets.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Interdisciplinary Applications
Hans Pinckaers, Wouter Bulten, Jeroen van der Laak, Geert Litjens
Summary: Prostate cancer is the most prevalent cancer among men in Western countries, and pathologists' evaluation is the gold standard for diagnosis. State-of-the-art convolutional neural networks are often patch-based and require detailed pixel-level annotations for effective training.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2021)
Article
Computer Science, Information Systems
James Wensel, Hayat Ullah, Arslan Munir
Summary: This paper proposes two transformer neural networks, Recurrent Transformer (ReT) and Vision Transformer (ViT), for human activity recognition. Extensive comparison experiments on four publicly available human action datasets show that the proposed ViT-ReT framework achieves significant improvements in both speed and accuracy, validating its suitability for human activity recognition in resource-constrained and real-time environments.
Article
Engineering, Biomedical
Jamie A. O'Reilly, Jordan Wehrman, Aaron Carey, Jennifer Bedwin, Thomas Hourn, Fawad Asadi, Paul F. Sowman
Summary: In this study, a computational model consisting of a three-dimensional convolutional neural network and a recurrent neural network was developed to investigate the sensitivity of the brain to facial event-related potentials (ERP). The model successfully simulated and predicted ERP waveforms in both simulated and real experiments. This approach has significant value in visual neuroscience research for studying the computational relationship between visual stimuli and neural activity.
JOURNAL OF NEURAL ENGINEERING
(2023)
Article
Biochemical Research Methods
Aniwat Juhong, Bo Li, Yifan Liu, Cheng-You Yao, Chia-Wei Yang, Dalen W. Agnew, Yu Leo Lei, Gary D. Luker, Harvey Bumpers, Xuefei Huang, Wibool Piyawattanametha, Zhen Qiu
Summary: Multispectral optoacoustic tomography (MSOT) is a useful technique for detailed analysis of biological samples, but it's time-consuming. This study proposes a deep learning model based on hybrid recurrent and convolutional neural networks to generate sequential cross-sectional images for MSOT. By using a larger step size for input images, the proposed model can generate additional high-resolution images, reducing acquisition time by approximately 71%.
JOURNAL OF BIOPHOTONICS
(2023)
Article
Computer Science, Information Systems
Tomasz Szandala
Summary: Technology has advanced rapidly in recent years, leading to the introduction of new solutions based on Machine Learning and Artificial Intelligence on a daily basis. However, understanding how these models make decisions has become challenging due to their complex black box decision-making process. Therefore, explainable artificial intelligence methods are crucial for further development. This paper discusses the need to revise existing state-of-the-art techniques in order to fully comprehend the prediction-generating process, and compares them with a new method called PRISM, which utilizes Principal Component Analysis for visualizing important features recognized by a given Convolutional Neural Network. The main objective of this paper is to examine how PRISM enhances the understanding of the decision-making process and introduce a tool called TorchPRISM for analyzing the output.
INFORMATION SCIENCES
(2023)