Article
Engineering, Mechanical
Kai Gao, Gang Liu, Wei Tang
Summary: The paper proposes an improved Manson-Halford (M-H) model with a load interaction factor, which can consider the load interaction effect and load sequence effect under multi-level loading. Experimental data from five materials verify the accuracy of the model under two-level and multi-level stress loadings.
INTERNATIONAL JOURNAL OF FATIGUE
(2021)
Article
Biology
Xiujuan Zhao, Yanping Zhang, Xiuquan Du
Summary: DFpin is a method for predicting protein-interacting nucleotides in RNA. It removes redundancy based on feature similarity and uses a deep forest model to extract key features, achieving an accuracy of 85.4%.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Biochemical Research Methods
Anup Kumar Halder, Soumyendu Sekhar Bandyopadhyay, Piyali Chatterjee, Mita Nasipuri, Dariusz Plewczynski, Subhadip Basu
Summary: This study proposes a multi-level feature-based approach for protein-protein interaction (PPI) prediction and introduces an improved evaluation strategy. Tested on six independent PPI datasets, the results show that this method outperforms state-of-the-art approaches in terms of prediction performance.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2022)
Article
Biology
Xingyue Gu, Yijie Ding, Pengfeng Xiao
Summary: Protein sequence classification is an important field in bioinformatics that plays a vital role in functional annotation, structure prediction, and understanding protein function and interactions. However, existing machine learning methods have limitations in terms of accuracy, precision, and generalization capabilities for different types of proteins. In this study, a protein sequence classifier called MLapRVFL is proposed, which incorporates Multi-Laplacian and L2,1-norm regularization to improve the model's generalization performance, robustness, and accuracy. Experimental results demonstrate that MLapRVFL outperforms popular machine learning methods and achieves superior predictive performance compared to previous studies. Overall, the proposed MLapRVFL method makes significant contributions to protein sequence prediction.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Mathematics
Xinghui Zhu, Liewu Cai, Zhuoyang Zou, Lei Zhu
Summary: Due to the low costs of storage and search, cross-modal retrieval hashing has attracted much interest in the big data era. Deep learning has significantly improved cross-modal representation capabilities, but existing methods cannot consider multi-label semantic learning and cross-modal similarity learning simultaneously. The proposed DMSFH method addresses these issues by using deep neural networks to extract cross-modal features and integrating multi-label semantic fusion to improve semantic discrimination learning. Additionally, graph regularization and pairwise loss help preserve the nearest neighbor relationship in Hamming subspace, resulting in competitive performance compared to state-of-the-art methods.
Article
Computer Science, Artificial Intelligence
Shaoru Guo, Yong Guan, Ru Li, Xiaoli Li, Hongye Tan
Summary: This paper introduces a Frame-based Multi-level Semantics Representation (FMSR) model, which utilizes frame knowledge and attention mechanisms to extract multi-level semantic information within sentences for text matching tasks. Experimental results show that the FMSR model outperforms state-of-the-art technologies in text matching tasks.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Tingjian Chen, Ying Zeng, Haoliang Yuan, Guo Zhong, Loi Lei Lai, Yuan Yan Tang
Summary: Unsupervised multi-view feature selection is an important research direction in pattern recognition and machine learning. This paper proposes a multi-level regularization-based unsupervised multi-view feature selection with adaptive graph learning, which has been proven to have superiority through experiments on multiple datasets.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2023)
Article
Computer Science, Information Systems
Xingjuan Cai, Yihao Cao, Yeqing Ren, Zhihua Cui, Wensheng Zhang
Summary: The study introduces a novel multi-objective evolutionary 3D face reconstruction model, which uses the feature map distortion algorithm to enhance network generalization ability and achieve better loss and NME values, demonstrating outstanding 3D facial reconstruction performance.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Lina Zhang, Runtao Yang, Defei Xia, Xiaorui Lin, Wanying Xiong
Summary: In this paper, a deep learning-based LPI prediction model called LPI-LSTM-ResNet is constructed, which effectively extracts features from LncRNAs and proteins and fuses them to improve LPI prediction performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Biochemical Research Methods
Satyajit Mahapatra, Vivek Raj Gupta, Sitanshu Sekhar Sahu, Ganapati Panda
Summary: In this paper, a novel hybrid approach combining deep neural network (DNN) and extreme gradient boosting classifier (XGB) is employed for predicting protein-protein interactions (PPI). The method achieves high accuracy in predicting both intra- and inter-species interactions, and has important implications for signaling pathway analysis and drug target prediction.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2022)
Article
Biochemical Research Methods
Mustafa Coskun, Mehmet Koyuturk
Summary: Using node similarity-based convolution matrices in GCNs can significantly improve the link prediction performance of GCN-based embeddings. Experimental results in biomedical networks show that this approach can enhance the performance of link prediction.
Article
Chemistry, Multidisciplinary
Iva Sovic, Josip Saric, Sinisa Segvic
Summary: Dense semantic forecasting, also known as anticipation of per-pixel semantics in a future unobserved frame, has been achieved using single-level regression methods but lacks consideration of skip connections. To address this, we propose warping shallow features from observed images with upsampled feature flow. Our training procedure enables recognition models to operate effectively with or without skip connections, revealing insights into the influence of skip connections on recognition accuracy. Our forecasting method achieves 70.2% mIoU 0.18 seconds into the future and 58.5% mIoU 0.54 seconds into the future, showing improved accuracy and promising directions for future work.
APPLIED SCIENCES-BASEL
(2023)
Review
Genetics & Heredity
Vladimir N. Uversky, Alessandro Giuliani
Summary: The text discusses the multi-level organization of nature, highlighting interactions and organization from the protein level to ecological systems. It challenges the traditional causative model based on genotype-phenotype distinction, proposing alternative top-down, bottom-up, and middle-out perturbation/control trajectories. The recent complex network studies reveal non-linear and non-bottom-up processes, shedding light on the deep nature of multi-level organization in biology.
FRONTIERS IN GENETICS
(2021)
Article
Biochemical Research Methods
Weihe Dong, Qiang Yang, Jian Wang, Long Xu, Xiaokun Li, Gongning Luo, Xin Gao
Summary: In this study, we proposed an innovative 'multi-modality attributes' learning-based framework for drug-protein interaction prediction. We utilized molecular transformer and graph convolutional networks to extract intermolecular sub-structural information and chemical semantic representations. By aggregating multiple biological representations, we learned condensed dimensional features of drugs, proteins, diseases, and side effects. The attribute representations were fused with adaptive learning weights to calculate the interaction score. Experimental results showed that our proposed method outperformed existing state-of-the-art frameworks in different experimental conditions.
BRIEFINGS IN BIOINFORMATICS
(2023)
Review
Computer Science, Artificial Intelligence
Gang Ren, Lei Diao, Jiyun Kim
Summary: Predicting the helpfulness of online reviews is crucial for consumers as it greatly influences their purchase decisions. To achieve accurate predictions, both textual and visual information in online reviews need to be considered. However, effectively fusing these heterogeneous modalities remains challenging. Therefore, we propose a novel method called DMFN, which utilizes multi-level information from both texts and images to enhance the representation of multimodal data. Our experiments on Yelp.com and Amazon.com datasets demonstrate that DMFN outperforms existing benchmark methods for Review Helpfulness Prediction (RHP).
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Paolo Campigotto, Stefano Teso, Roberto Battiti, Andrea Passerini
Summary: CLEO is a preference elicitation algorithm that recommends optimal configurations in hybrid domains through interactive optimization, incorporating Max-SMT technology and learning to rank implementation, adapting to uncertainty in the decision maker's utility and noisy feedback.
ARTIFICIAL INTELLIGENCE
(2021)
Article
Mathematics, Interdisciplinary Applications
Wanyi Zhang, Qiang Shen, Stefano Teso, Bruno Lepri, Andrea Passerini, Ivano Bison, Fausto Giunchiglia
Summary: Various studies have investigated the predictability of different aspects of human behavior such as mobility patterns, social interactions, and shopping and online behaviors. The key assumption is that human behavior is deliberated based on an individual's own perception of the situation. Contextual dimensions like time, location, activity, and social ties play a significant role in the predictability of individuals' behaviors, with multi-modality information being crucial for accurate predictions.
Article
Computer Science, Artificial Intelligence
Antonio Longa, Giulia Cencetti, Bruno Lepri, Andrea Passerini
Summary: This study introduces a technique based on egocentric temporal neighborhoods to extract temporal motifs in temporal networks, bypassing the graph isomorphism problem and enabling the algorithm to mine larger motifs. By focusing on the temporal dynamics of interactions of specific nodes, the method allows for the extraction of interpretable temporal motifs.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Computer Science, Artificial Intelligence
Wanyi Zhang, Mattia Zeni, Andrea Passerini, Fausto Giunchiglia
Summary: Mobile Crowd Sensing (MCS) integrates sensor data and user-generated content in the IoT paradigm. The poor quality of human-provided content, mainly due to inaccurate input, is a major challenge. This paper presents Skeptical Learning, an algorithm that checks and fixes user feedback to address this issue. The results demonstrate the algorithm's advantages in dealing with mislabeling problems and improving prediction performance.
Article
Computer Science, Artificial Intelligence
Andrea Bontempelli, Fausto Giunchiglia, Andrea Passerini, Stefano Teso
Summary: Knowledge drift is a special form of concept drift that occurs in hierarchical classification, where changes in concept vocabulary, distributions, and relations can affect classification accuracy. Identifying the type of knowledge drift is crucial for improving classifier performance, and involving users in interactive disambiguation can lead to significant enhancements in prediction accuracy.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Computer Science, Artificial Intelligence
Mohit Kumar, Samuel Kolb, Stefano Teso, Luc De Raedt
Summary: Combinatorial optimisation problems are common in artificial intelligence. This study introduces a novel approach for learning these problems from contextual examples using the MAX-SAT formalism. Theoretical results show that high-quality MAX-SAT models can be learned from contextual examples, as long as the data satisfies a representativeness condition.
ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Giovanni De Toni, Bruno Lepri, Andrea Passerini
Summary: This paper proposes a new approach that learns a program to generate explainable counterfactual actions, aiming to provide explanations on how to change unfavourable decisions made by a black-box machine learning model. The approach leverages program synthesis techniques, reinforcement learning with Monte Carlo Tree Search, and rule learning to extract explanations for each recommended action. Experimental evaluation shows that the proposed approach, FARE (eFficient counterfActual REcourse), generates effective interventions with significantly fewer queries to the black-box classifier compared to existing solutions, while providing interpretable explanations.
Article
Psychology, Experimental
Stefano Fait, Stefania Pighin, Andrea Passerini, Francesco Pavani, Katya Tentori
Summary: Bayesianism assumes that probabilistic updating is independent of information modality. This study investigates whether probability judgments based on visual and auditory information conform to this assumption. The results show that when information is acquired through a single modality, probabilistic updating is consistent with Bayesianism. However, when information comes from multiple modalities, there is a synergy-contrast effect where judgments are more accurate when visual and auditory information individually and jointly support the hypothesis than when they support different hypotheses.
Proceedings Paper
Computer Science, Information Systems
Alessia Bertugli, Stefano Vincenzi, Simone Calderara, Andrea Passerini
Summary: FUSION is a learning strategy that enables a neural network to learn quickly and continually on unlabelled data streams and unbalanced tasks, maximizing knowledge extraction and utilizing supervised information when available.
MACHINE LEARNING, OPTIMIZATION, AND DATA SCIENCE, LOD 2022, PT I
(2023)
Article
Mathematics, Interdisciplinary Applications
Nicolo Alessandro Girardini, Simone Centellegher, Andrea Passerini, Ivano Bison, Fausto Giunchiglia, Bruno Lepri
Summary: This study analyzes the changes in students' behavior during the COVID-19 pandemic. By comparing the qualitative and quantitative differences in their daily routines between 2018 and 2020, the study finds that despite restrictions, there are minimal changes in the activities performed by students, but adaptation primarily occurs in the location and sociality dimensions.
Review
Microbiology
Francesco Asnicar, Andrew Maltez Thomas, Andrea Passerini, Levi Waldron, Nicola Segata
Summary: This article reviews the importance and applications of machine learning in microbiology, including tasks such as predicting antibiotic resistance and associating with host diseases. It provides a basic toolbox for microbiologists to understand and apply machine learning.
NATURE REVIEWS MICROBIOLOGY
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Emanuele Marconato, Gianpaolo Bontempo, Stefano Teso, Elisa Ficarra, Simone Calderara, Andrea Passerini
Summary: This paper analyzes the issue of Catastrophic Forgetting in Deep Learning models during continual learning and investigates the effect of high-level feature supervision on this problem. Two metrics are introduced to evaluate the loss of information on learned concepts as new experiences are encountered. Additionally, the paper shows that saliency maps remain more stable with attribute supervision.
IMAGE ANALYSIS AND PROCESSING, ICIAP 2022 WORKSHOPS, PT II
(2022)
Article
Ethics
Mirco Nanni, Gennady Andrienko, Albert-Laszlo Barabasi, Chiara Boldrini, Francesco Bonchi, Ciro Cattuto, Francesca Chiaromonte, Giovanni Comande, Marco Conti, Mark Cote, Frank Dignum, Virginia Dignum, Josep Domingo-Ferrer, Paolo Ferragina, Fosca Giannotti, Riccardo Guidotti, Dirk Helbing, Kimmo Kaski, Janos Kertesz, Sune Lehmann, Bruno Lepri, Paul Lukowicz, Stan Matwin, David Megias Jimenez, Anna Monreale, Katharina Morik, Nuria Oliver, Andrea Passarella, Andrea Passerini, Dino Pedreschi, Alex Pentland, Fabio Pianesi, Francesca Pratesi, Salvatore Rinzivillo, Salvatore Ruggieri, Arno Siebes, Vicenc Torra, Roberto Trasarti, Jeroen van den Hoven, Alessandro Vespignani
Summary: The rapid spread of COVID-19 requires quick and effective tracking of virus transmission chains and early detection of outbreaks, especially as lockdown measures are lifted. A decentralized approach to contact-tracing apps offers better protection of citizens' privacy and allows for detailed information gathering for infected individuals in a privacy-preserving manner, enabling more effective contact tracing and early detection of outbreak hotspots.
ETHICS AND INFORMATION TECHNOLOGY
(2021)