Article
Chemistry, Analytical
Ariyo Oluwasanmi, Muhammad Umar Aftab, Edward Baagyere, Zhiguang Qin, Muhammad Ahmad, Manuel Mazzara
Summary: This study proposes three artificial intelligence models for analyzing and detecting anomalies in human heartbeat signals using deep learning algorithms. The models include an attention autoencoder, a variational autoencoder, and a long short-term memory network. The three models exhibit outstanding ability in detecting healthy heartbeats in patients with severe congestive heart failure.
Article
Computer Science, Information Systems
Imayanmosha Wahlang, Arnab Kumar Maji, Goutam Saha, Prasun Chakrabarti, Michal Jasinski, Zbigniew Leonowicz, Elzbieta Jasinska
Summary: This article experiments with deep learning methodologies in echocardiogram, focusing on classification of normal and abnormal conditions as well as different types of regurgitation. The use of videographic images distinguishes this work from existing methods, showing that deep learning methodologies outperform SVM in normal or abnormal classification. VAE performs better with static images, while LSTM is more effective with videographic images.
Article
Environmental Sciences
Shaoqi Yu, Xiaorun Li, Shuhan Chen, Liaoying Zhao
Summary: Neural network-based anomaly detection methods have gained attention for their reconstruction ability in hyperspectral remote sensing. A novel probability distribution representation detector (PDRD) is proposed to explore intrinsic distributions of backgrounds and anomalies. Experimental results show the method's accuracy and efficiency compared to state-of-the-art detection methods.
Article
Computer Science, Artificial Intelligence
Xinyu Chen, Jiajie Xu, Rui Zhou, Wei Chen, Junhua Fang, Chengfei Liu
Summary: This paper focuses on trajectory generation problem and proposes two advanced solutions, TrajGAN and TrajVAE, which utilize LSTM, GAN, and VAE frameworks to model and generate trajectories. The accuracy and stability of the methods are validated through multiple trajectory similarity metrics in several experiments.
Article
Automation & Control Systems
Luyue Lin, Bo Liu, Xin Zheng, Yanshan Xiao
Summary: This article introduces a classifier method based on a variational autoencoder (CFVAE) to improve the performance of CNN learning with insufficient samples. The method utilizes a prior classifier to generate label and latent variable distribution information, and a posterior classifier to enhance latent variables for improved performance. Experiments demonstrate that the proposed CFVAE outperforms other methods in accuracy.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Computer Science, Hardware & Architecture
Georgios Zioviris, Kostas Kolomvatsos, George Stamoulis
Summary: The banking sector is undergoing a serious transformation driven by the application of artificial intelligence (AI). This paper introduces a novel multistage deep learning model that combines autoencoders for feature selection and deep convolutional neural network for fraud detection in transaction streams.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Automation & Control Systems
Xiuli Zhu, Seshu Kumar Damarla, Kuangrong Hao, Biao Huang
Summary: This article introduces the application of data-driven soft sensors in industrial processes and proposes improved methods to enhance accuracy and stability. The effectiveness of the proposed methods is validated through a case study on polyester polymerization process.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Lin Yang, Wentao Fan, Nizar Bouguila
Summary: This article proposes a clustering method based on variational autoencoder with spherical latent embeddings. The method improves clustering accuracy by using the von Mises-Fisher mixture model prior and a dual VAE structure, and enhances model robustness through an augmented loss function.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Chemistry, Analytical
Jie Jiang, Yin Zou, Lidong Chen, Yujie Fang
Summary: In this paper, an unsupervised hierarchical indoor localization framework combining unsupervised network variational autoencoder and visual-based SfM approach is proposed to extract global and local features for precise image localization and pose estimation. By using global features for image retrieval at the level of scene map and subsequently estimating pose through 2D-3D matches, the proposed method achieves promising results in accuracy and efficiency.
Article
Microbiology
Ying Feng, Moutong Chen, Xianhu Wei, Honghui Zhu, Jumei Zhang, Youxiong Zhang, Liang Xue, Lanyan Huang, Guoyang Chen, Minling Chen, Yu Ding, Qingping Wu
Summary: Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry fingerprinting, combined with pseudotargeted metabolomics and deep learning, enables rapid and accurate identification and visualization of pathogens.
FRONTIERS IN MICROBIOLOGY
(2022)
Article
Computer Science, Information Systems
Haomiao Yang, Mengyu Ge, Kunlan Xiang, Xuejun Bai, Hongwei Li
Summary: This article proposes FedVAE, an FL framework based on variational autoencoder (VAE) for remote patient monitoring. FedVAE consists of two lightweight VAEs, one for reducing communication overhead and slow convergence rate caused by non-IID data, and the other for eliminating training bias through generating minority class samples. Experimental results show that FedVAE achieves an AUC value of 0.9937, higher than the traditional centralized model (0.9931). Fine-tuning the global model with personalization raises the average AUC to 0.9947. Moreover, FedVAE outperforms vanilla FL by improving AUC by 0.87% while reducing communication traffic by at least 95%.
IEEE SYSTEMS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Shuo Wang, Shangyu Chen, Tianle Chen, Surya Nepal, Carsten Rudolph, Marthie Grobler
Summary: This study proposes an adversarial attack strategy that implements fine-granularity, semantic-meaning-oriented structural perturbations on images. The method manipulates the semantic attributes of images through the use of disentangled latent codes to construct adversarial perturbations. The empirical evaluations demonstrate the strong attack capability of this method against black-box classifiers and establish the existence of a universal semantic adversarial example.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Multidisciplinary Sciences
Mohamed R. Ibrahim, James Haworth, Aldo Lipani, Nilufer Aslam, Tao Cheng, Nicola Christie
Summary: Modeling the global spread of coronavirus and learning trends at global and country levels is crucial for fighting the pandemic. A novel variational-LSTM Autoencoder model is introduced to predict coronavirus spread globally, incorporating historical data, urban characteristics, and governmental measures. The method also adjusts spatial dependencies among countries while forecasting the spread. Trained models show high validation for short and long-term spread forecasts, providing a useful tool for decision and policymaking worldwide.
Article
Computer Science, Artificial Intelligence
Ruochen Li, Nannan Li, Wenmin Wang
Summary: This article introduces a method to improve the performance of audio-visual event retrieval by simulating the processing function of the human brain. The proposed InfoIIM network enhances feature representation and alignment, and the InfoMax-VAE model improves feature learning and intra-modal retrieval performance. The effectiveness of the method is verified on the AVE dataset, showing superior performance compared to existing algorithms. Future research directions are also suggested to inspire relevant researchers.
INTERNATIONAL JOURNAL OF MULTIMEDIA INFORMATION RETRIEVAL
(2023)
Article
Chemistry, Multidisciplinary
Jaime Carracedo-Cosme, Carlos Romero-Muniz, Ruben Perez
Summary: Despite the resolution provided by AFM with CO-functionalized, identifying molecular systems based solely on AFM images remains challenging. This study presents a deep learning model trained on a theoretically generated dataset for automatic classification of AFM experimental images. The study explores the limitations of standard pattern recognition models and proposes a model with optimal depth for accurate results and generalization ability. A variational autoencoder is shown to efficiently incorporate characteristic features from very few experimental images into the training set, ensuring high accuracy in classification.
