Article
Computer Science, Artificial Intelligence
Shao-Yu Yin, Yu Huang, Tien-Yu Chang, Shih-Fang Chang, Vincent S. Tseng
Summary: Continual learning is an emerging research branch of deep learning that aims to continuously learn a model for a series of tasks without forgetting knowledge obtained from previous tasks. In this paper, the authors propose a novel method called Temporal Teacher Distillation (TTD) based on attentive recurrent neural networks to address the problem of catastrophic forgetting in temporal-based continual learning. Experimental results show that the proposed TTD significantly outperforms state-of-the-art methods in terms of accuracy and forgetting measures.
Article
Chemistry, Analytical
Ke Zang, Wenqi Wu, Wei Luo
Summary: This paper introduces a neural network parameter optimization method based on sparse learning, which can reduce parameters without compromising network performance. Experimental results demonstrate that the proposed method effectively reduces model parameters and improves network generalization ability.
Article
Geochemistry & Geophysics
Weilian Zhou, Sei-ichiro Kamata, Zhengbo Luo, Haipeng Wang
Summary: This study proposes a novel multiscanning strategy with recurrent neural networks (RNN) for hyperspectral image (HSI) classification. By considering different scanning orders and concatenating features as input to the RNN, the network can effectively capture the spatial dependence of HSI patches and achieve competitive classification performance on three datasets.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Review
Environmental Sciences
Leiyu Chen, Shaobo Li, Qiang Bai, Jing Yang, Sanlong Jiang, Yanming Miao
Summary: This article summarizes the application of deep learning in image classification, covering the development of CNNs from their predecessors to the latest network architectures, as well as a comprehensive comparison and analysis of various image classification methods.
Article
Computer Science, Artificial Intelligence
Samina Amin, Abdullah Alharbi, M. Irfan Uddin, Hashem Alyami
Summary: The study analyzed public discussions on COVID-19 symptoms on Twitter using machine and deep learning models, showing that optimizing AUC can improve model performance, and LSTM has the highest accuracy in detecting COVID-19 symptoms.
Article
Computer Science, Information Systems
Angel Lopez Garcia-Arias, Yasuyuki Okoshi, Masanori Hashimoto, Masato Motomura, Jaehoon Yu
Summary: Accurate neural networks can be found by pruning a randomly initialized overparameterized model, which eliminates the need for weight optimization. The resulting subnetworks are small, sparse, and ternary, making them ideal for efficient hardware implementation. However, the challenge lies in finding optimal connectivity patterns.
Article
Biochemical Research Methods
Amelia Villegas-Morcillo, Angel M. Gomez, Juan A. Morales-Cordovilla, Victoria Sanchez
Summary: The paper introduces a deep learning architecture to address the protein fold recognition task by processing protein residue-level features. By combining 1D-convolutional layers with gated recurrent unit (GRU) layers, the model copes with issues related to highly variable protein sequence lengths and extracts fold-related embeddings of fixed size for each protein domain. These embeddings outperform other methods significantly, especially at the fold level, according to evaluation results on various datasets.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2021)
Article
Engineering, Multidisciplinary
Youssef Ghatas, Magda Fayek, Mayada Hadhoud
Summary: This study investigates deep learning approaches to solve the problem of musical difficulty estimation. By converting symbolic music into piano roll representation and training convolutional neural networks, the proposed hybrid deep model achieves a significant improvement compared to previous studies, with an F1 score of 76.26%.
ALEXANDRIA ENGINEERING JOURNAL
(2022)
Article
Environmental Sciences
Tiago Marto, Alexandre Bernardino, Goncalo Cruz
Summary: This work proposes an active learning methodology for the segmentation of fire and smoke in video images. The model learns incrementally over several active learning rounds, selecting informative samples to update the training set. Using active learning in classification and segmentation tasks resulted in improved accuracy and mean intersection over union by 2%, while achieving similar results to non-active learning with fewer labeled data samples.
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
Computer Science, Artificial Intelligence
Jianxiong Tang, Jian-Huang Lai, Xiaohua Xie, Lingxiao Yang, Wei-Shi Zheng
Summary: This paper proposes a fast and memory-efficient Activation Consistency Coupled ANN-SNN (AC2AS) framework for training SNN in low-power environments. The framework utilizes a weight-shared architecture between ANN and SNN, as well as spiking mapping units, to achieve fast training and ensure activation consistency for SNN. Experimental results show that AC2AS-based models perform well on benchmark datasets and achieve comparable accuracy with reduced time steps, training time, GPU memory costs, and spike activities compared to the Spike-based BP model.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Ricardo A. A. Borsoi, Tales Imbiriba, Pau Closas
Summary: Multitemporal hyperspectral unmixing (MTHU) is a key tool for analyzing hyperspectral image sequences, as it reveals the dynamic evolution of endmembers and their abundances in a scene. In this study, an unsupervised MTHU algorithm based on variational recurrent neural networks is proposed, which addresses the challenge of accounting for the spatial and temporal variability of endmembers in an unsupervised manner. The algorithm employs a stochastic model, a low-dimensional parametrization model, and deep variational inference to accurately estimate the abundances and endmembers.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Engineering, Civil
Wladimir M. Gonzalez, Andres Ferrada, Ruben L. Boroschek, Enrique Lopez Droguett
Summary: This article focuses on the application of Recurrent Neural Networks (RNN) with Long-Short Term Memory (LSTM) blocks to modal tracking in medium-rise buildings. The models are trained to characterize the environmental trend in the modal frequency to detect changes of state or damage. The models perform well in capturing the annual evolution of the modal frequency and are suitable for damage detection.
ENGINEERING STRUCTURES
(2022)
Article
Computer Science, Information Systems
Gheith A. Abandah, Mohammed Z. Khedher, Mohammad R. Abdel-Majeed, Hamdi M. Mansour, Salma F. Hulliel, Lara M. Bisharat
Summary: This paper proposes a solution to classify input Arabic text into 16 poetry meters and prose, and investigates the automatic diacritization of Arabic poetry. Machine learning approach is adopted using neural networks to achieve high accuracy in classifying and diacritizing Arabic poetry.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Monika Jyotiyana, Nishtha Kesswani, Munish Kumar
Summary: Deep learning techniques have great potential in healthcare, offering fast and accurate classification and prediction of diseases. This study proposes a deep neural network-based classification model for Parkinson's disease with an accuracy of 94.87%.
Article
Computer Science, Artificial Intelligence
Roldano Cattoni, Mattia Antonino Di Gangi, Luisa Bentivogli, Matteo Negri, Marco Turchi
Summary: End-to-end spoken language translation (SLT) has gained popularity recently, but faces challenges due to the scarcity of publicly available corpora. MuST-C is a large multilingual speech translation corpus built from English TED Talks, with unique features including language coverage, size, variety of topics, and data quality.
COMPUTER SPEECH AND LANGUAGE
(2021)
Article
Computer Science, Interdisciplinary Applications
Giovanni Giacalone, Marco Barra, Angelo Bonanno, Gualtiero Basilone, Ignazio Fontana, Monica Calabro, Simona Genovese, Rosalia Ferreri, Giuseppa Buscaino, Salvatore Mazzola, Riko Noormets, Christopher Nuth, Giosue Lo Bosco, Riccardo Rizzo, Salvatore Aronica
Summary: In this study, acoustic data collected in Kongsfjorden, Svalbard, was analyzed to develop a method for identifying and classifying fish aggregations using 3D acoustic patterns. The results suggest that three distinct groups can be identified mathematically. This approach shows promise for improving monitoring programs for marine resources and can be applied to climate change research.
ENVIRONMENTAL MODELLING & SOFTWARE
(2022)
Article
Computer Science, Interdisciplinary Applications
Ignazio Fontana, Marco Barra, Angelo Bonanno, Giovanni Giacalone, Riccardo Rizzo, Olga Mangoni, Simona Genovese, Gualtiero Basilone, Rosalia Ferreri, Salvatore Mazzola, Giosue Lo Bosco, Salvatore Aronica
Summary: Acoustic surveys play a significant role in assessing the distribution and abundance of pelagic organisms. The identification of species in acoustic observations is usually based on biological sampling and expert knowledge. This study examines the use of unsupervised clustering methods for identifying krill species and finds that k-means performs better than hierarchical methods. The findings highlight the importance of selecting specific variables for clustering analysis to improve accuracy.
ENVIRONMENTAL MODELLING & SOFTWARE
(2022)
Article
Biochemistry & Molecular Biology
Federica Scalia, Rosario Barone, Francesca Rappa, Antonella Marino Gammazza, Fabrizio Lo Celso, Giosue Lo Bosco, Giampaolo Barone, Vincenzo Antona, Maria Vadala, Alessandra Maria Vitale, Giuseppe Donato Mangano, Domenico Amato, Giusy Sentiero, Filippo Macaluso, Kathryn H. Myburgh, Everly Conway de Macario, Alberto J. L. Macario, Mario Giuffre, Francesco Cappello
Summary: Recognition of diseases associated with mutations of the chaperone system genes (chaperonopathies) is increasing, but the impact of the mutation on the chaperone molecule and the mechanisms underlying tissue abnormalities are not well understood. This study examined the histological features of skeletal muscle from a patient with a severe, early onset, distal motor neuropathy carrying a mutation on the CCT5 subunit (MUT). The mutated muscle showed significant modifications including fiber atrophy, disruption of tissue architecture, and apoptosis. The study also found abnormal localization and precipitation of various proteins in the mutated muscle. In silico analyses of the mutant CCT5 molecule revealed abnormalities that could impair chaperoning functions. Further in vitro and in vivo analysis of the mutated CCT5 is anticipated to provide additional insights on subunit involvement in neuromuscular disorders.
FRONTIERS IN MOLECULAR BIOSCIENCES
(2022)
Article
Computer Science, Software Engineering
Domenico Amato, Giosue Lo Bosco, Raffaele Giancarlo
Summary: Learned Indexes use a model to restrict the search range of a sorted table, and using the SOSD benchmarking software, this study demonstrates that k-ary search is more efficient in certain computer architectures. This research provides guidelines for selecting the search routine within the learned indexing framework.
SOFTWARE-PRACTICE & EXPERIENCE
(2023)
Article
Biochemistry & Molecular Biology
Claudia Musial, Narcyz Knap, Renata Zaucha, Paulina Bastian, Giampaolo Barone, Giosue Lo Bosco, Fabrizio Lo-Celso, Lucyna Konieczna, Mariusz Belka, Tomasz Baczek, Antonella Marino Gammazza, Alicja Kuban-Jankowska, Francesco Cappello, Stephan Nussberger, Magdalena Gorska-Ponikowska
Summary: Research shows that 2-ME can inhibit the growth of lung cancer cells and increase protein palmitoylation and oxidative stress phenomena. Through metabolomics analysis of blood serum, 2-ME can be used as a selective tumor biomarker in specific types of lung cancer. In addition, computational analysis suggests that 2-ME has relatively safe bioactivity in healthy human cells.
Proceedings Paper
Computer Science, Information Systems
Salvatore Calderaro, Giosue Lo Bosco, Riccardo Rizzo, Filippo Vella
Summary: This study proposes a histopathological image classification method based on convolutional neural networks. By utilizing metric learning, the network learns a representation that clusters labeled samples based on their characteristics, improving classification performance and supporting labeling decisions.
2022 IEEE EIGHTH INTERNATIONAL CONFERENCE ON MULTIMEDIA BIG DATA (BIGMM 2022)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Antonino D'Alessandro, Andrea Di Benedetto, Giosue Lo Bosco, Anna Figlioli
Summary: This work introduces an active learning approach to improve the classification of seismo-volcanic events, particularly explosion quakes, using a random forest classifier. Human intervention is involved to annotate uncertain data, resulting in improved probability distribution of events after intervention.
2022 IEEE CONFERENCE ON EVOLVING AND ADAPTIVE INTELLIGENT SYSTEMS (IEEE EAIS 2022)
(2022)
Proceedings Paper
Engineering, Electrical & Electronic
Mariella Farella, Giuseppe Chiazzese, Giosue Lo Bosco
Summary: This paper proposes the design of an avatar system with question-answering capabilities for immersive navigation of cultural heritage sites. The system utilizes technologies like Virtual Reality, Augmented Reality, and Artificial Intelligence to enhance user experience.
2022 IEEE 21ST MEDITERRANEAN ELECTROTECHNICAL CONFERENCE (IEEE MELECON 2022)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Gabriella Casalino, Alfredo Cuzzocrea, Giosue Lo Bosco, Mariano Maiorana, Giovanni Pilato, Daniele Schicchi
Summary: Satire is a form of criticizing individuals or ideas through ridicule, often used to denounce political and societal issues. Automatic detection of satire is a challenging task, but effective deep learning architectures show promise in addressing this issue.
FLEXIBLE QUERY ANSWERING SYSTEMS (FQAS 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Angelo Luigi Megna, Daniele Schicchi, Giosue Lo Bosco, Giovanni Pilato
Summary: Researchers have developed a deep learning-based system for simplifying Italian texts, inspired by a state of the art system for the English language. Through training and testing on the Italian version of Newsela, the system has shown promising results with a SARI value of 30.17.
2021 IEEE 15TH INTERNATIONAL CONFERENCE ON SEMANTIC COMPUTING (ICSC 2021)
(2021)
Letter
Infectious Diseases
Antonella Marino Gammazza, Sebastien Legare, Giosue Lo Bosco, Alberto Fucarino, Francesca Angileri, Massimiliano Oliveri, Francesco Cappello
Article
Biochemistry & Molecular Biology
Claudia Musial, Narcyz Knap, Renata Zaucha, Paulina Bastian, Giampaolo Barone, Giosue Lo Bosco, Fabrizio Lo-Celso, Lucyna Konieczna, Mariusz Belka, Tomasz Baczek, Antonella Marino Gammazza, Alicja Kuban-Jankowska, Francesco Cappello, Stephan Nussberger, Magdalena Gorska-Ponikowska
Summary: 2-Methoxyestradiol (2-ME) as an inhibitor for non-small cell lung cancer cells may serve as a potential therapeutic approach, reducing cell viability, promoting protein palmitoylation and oxidative stress, and showing relative safety in healthy human cells compared to other estrogen metabolism intermediates.