Article
Engineering, Civil
Xing Liu, Cheng Qian, Wei Yu, David Griffith, Avi Gopstein, Nada Golmie
Summary: In this paper, the authors propose a deep reinforcement learning-based approach to automate network configurations in dynamic network environments such as the Internet of Vehicles (IoV). They use a collection of neural networks to convert the observations of a communication environment into key features and then train a deep Q neural network (DQN) to select optimal network configurations for vehicles in the IoV environment. They also consider both centralized and distributed training strategies and evaluate the efficacy of their approach using an IoV simulation platform.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Qijin Wang, Yu Qian, Yating Hu, Chao Wang, Xiaodong Ye, Hongqiang Wang
Summary: Object detection under one-level feature is difficult and YOLOF solves some problems with object scale and sample quantity balance. To improve performance, especially for smaller objects, a new object detector called M2YOLOF is proposed with attention-based encoder and dynamic sample selection policy. M2YOLOF strengthens feature maps' contextual details and balances the rationality of training samples. Experimental results show the effectiveness of our method.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Multidisciplinary Sciences
Hassan Ashraf, Asim Waris, Muhammad Fazeel Ghafoor, Syed Omer Gilani, Imran Khan Niazi
Summary: This study proposes a novel deep learning-based, fully automated approach to skin lesion segmentation, including sophisticated pre and postprocessing approaches. The proposed method achieved high segmentation accuracy with the use of three deep learning models, test time augmentation, and conditional random field. It showed promising results on different datasets, demonstrating its scalability and robustness.
SCIENTIFIC REPORTS
(2022)
Article
Multidisciplinary Sciences
Asad Khattak, Zartashia Mehak, Hussain Ahmad, Muhammad Usama Asghar, Muhammad Zubair Asghar, Aurangzeb Khan
Summary: Customer churn is a problem for businesses to retain customers, and machine learning and deep learning techniques have been used to identify customer churn. However, previous research has shown unexpected results when using machine learning classifiers and traditional feature encoding methods. In this study, a hybrid deep learning model called BiLSTM-CNN is proposed to predict customer churn effectively and improve accuracy.
SCIENTIFIC REPORTS
(2023)
Article
Engineering, Civil
Liwen Xing, Ningbo Cui, Li Guo, Taisheng Du, Daozhi Gong, Cun Zhan, Long Zhao, Zongjun Wu
Summary: A novel hybrid deep learning model combining deep belief network (DBN) and long short-term memory (LSTM) modules was developed for ET0 estimation using meteorological data on the Loess Plateau, and it showed the best performance in RH-based, Rn-based, and T-based models.
JOURNAL OF HYDROLOGY
(2022)
Article
Green & Sustainable Science & Technology
He Jun
Summary: Environmental toxic reduction (ETR) is important for conducting risk assessments on chemical goods and effluents. Toxic waste can harm humans, animals, and plants, and deep learning can be used to predict the toxicity of chemicals.
SUSTAINABLE ENERGY TECHNOLOGIES AND ASSESSMENTS
(2022)
Article
Computer Science, Information Systems
Priya A. G. Varshini, Anitha K. Kumari, Vijayakumar Varadarajan
Summary: This paper explores different methods for software project estimation, comparing single model approaches with ensemble techniques. Results indicate that stacking using random forest outperforms other methods including single models and ensemble techniques.
Article
Medicine, General & Internal
Eun-Gyeong Kim, Il-Seok Oh, Jeong-Eun So, Junhyeok Kang, Van Nhat Thang Le, Min-Kyung Tak, Dae-Woo Lee
Summary: This study proposes the use of deep learning models for estimating cervical vertebral maturation from lateral cephalograms, and introduces a stepwise segmentation-based model. The results show that the three-step segmentation-based model achieved the best accuracy compared to non-segmentation-based models.
JOURNAL OF CLINICAL MEDICINE
(2021)
Article
Chemistry, Physical
Jakub K. Sowa, Sean T. Roberts, Peter J. Rossky
Summary: Semiconducting nanocrystals passivated with organic ligands are a powerful platform for light harvesting, light-driven chemical reactions, and sensing. This study develops a machine-learned force field trained on DFT data to investigate the surface chemistry of a PbS nanocrystal with acetate ligands. The study demonstrates that carboxylate ligands passivate metal-rich surfaces through a wide range of geometries and explores the corresponding ligand IR spectrum. This work illustrates the potential of machine-learned force fields in computational modeling of these materials.
JOURNAL OF PHYSICAL CHEMISTRY LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
John Taco, Pradeep Kundu, Jay Lee
Summary: Deep learning is a flexible technique for pattern recognition in raw sensor data, but it requires complex optimization and high computational resources. Classical machine learning algorithms rely on domain knowledge and feature engineering, which may not always be available in industry. To overcome these limitations, a method based on the deep forest algorithm is proposed, which automatically learns characteristics from multivariate time series data and achieves faster and more accurate failure mode classification.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Medicine, General & Internal
Abdul Rahaman Wahab Sait
Summary: Diabetic retinopathy is a severe complication of diabetes that affects a large portion of the population in Saudi Arabia. Existing systems for treating diabetic retinopathy patients are computationally expensive and may produce false positive outcomes due to imbalanced datasets. To address these issues, the author developed a lightweight deep-learning model that utilized image pre-processing and feature extraction techniques to predict the severity of diabetic retinopathy. The proposed model achieved high accuracy and low computational complexity, making it suitable for use in healthcare centers and as a mobile application for supporting clinicians. The author plans to further improve the model's efficiency in detecting diabetic retinopathy from low-quality fundus images.
Article
Computer Science, Artificial Intelligence
Shakeel Ahmad, Muhammad Zubair Asghar, Fahad Mazaed Alotaibi, Yasir D. Alotaibi
Summary: The World Health Organization reports that cardiovascular disease is the leading cause of death, accounting for 31% of global deaths. A study proposes using a deep learning model, specifically a CNN with bidirectional long/short-term memory, to efficiently predict cardiovascular disease from patient data. Experimental outcomes show that this hybrid deep learning technique achieves an accuracy of 94.507% in predicting cardiovascular illness.
Article
Robotics
Daniel Mox, Vijay Kumar, Alejandro Ribeiro
Summary: The research team proposed a data-driven approach using convolutional neural networks to optimize the algebraic connectivity of robot teams. This method has the potential for online applications and is significantly faster than traditional optimization-based approaches.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Computer Science, Information Systems
Yueh-Peng Chen, Tzuo-Yau Fan, Her-Chang Chao
Summary: This study introduces a watermark copyright identification model based on deep learning technology called WMNet, which uses a large dataset of distorted watermarks to accurately identify copyright information. By categorizing watermarks in the training dataset, the model is able to improve the effectiveness of copyright verification.
Article
Environmental Sciences
Yun Chen, Juan Guerschman, Yuri Shendryk, Dave Henry, Matthew Tom Harrison
Summary: This study successfully developed a sequential neural network model for estimating pasture biomass in five dairy farms in northern Tasmania, Australia, using high-resolution imagery and advanced machine learning techniques, resulting in a relatively high prediction accuracy. The research also revealed the impact of in situ measurements, pasture management, and grazing practices on the model's predictions on different farms.