Article
Geochemistry & Geophysics
Shuaiyuan Du, Kewei Wang, Zhiguo Cao
Summary: Most current infrared small target detection methods attempt to fuse local and global information by using single-scale inputs and creating a multiscale feature pyramid during network feeding forward. Our research finds that using high-resolution inputs can improve recall, while low-resolution inputs improve precision. To address these issues, we propose BPR-Net, an approach that balances precision and recall via a novel multiscale attention mechanism.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Pasi Franti, Radu Mariescu-Istodor
Summary: Precision and recall, classical measures in machine learning, are based on exact matching. However, in pattern recognition, inexact matching is often preferred. To address this issue, soft variants of precision and recall are introduced, based on application-specific similarity measures.
PATTERN RECOGNITION LETTERS
(2023)
Article
Remote Sensing
Joni Storie, C. D. Storie, C. J. Henry, M. Sokolov, B. Murray, J. Cameron, R-M Tsenov
Summary: Estimating surface water using satellite data can improve predictions for hydroelectric energy generation, but different threshold algorithms have varying levels of precision and recall in different seasons. Validation data limitations hinder accuracy assessment of these methods.
REMOTE SENSING LETTERS
(2021)
Article
Mathematics
Jose A. Saez, Jose L. Romero-Bejar
Summary: Real-world classification data often contain noise, which can affect the accuracy and complexity of models. Building ensembles of classifiers, such as bagging, has shown potential in reducing the effects of noise. This paper investigates the usage of bagging techniques in complex problems where noise impacts decision boundaries among classes, and finds that bagging can achieve better accuracy and robustness in the presence of borderline noise.
Article
Chemistry, Multidisciplinary
Iurii Katser, Viacheslav Kozitsin, Victor Lobachev, Ivan Maksimov
Summary: Offline changepoint detection algorithms are used for optimal signal segmentation, based on known statistical properties and appropriate models. While ensemble approaches are known for increasing robustness and dealing with challenges, they are less formalized in CPD problems. The proposed unsupervised CPD ensemble procedure outperforms non-ensemble procedures in numerical experiments, with a focus on analyzing common CPD algorithms, scaling, and aggregation functions.
APPLIED SCIENCES-BASEL
(2021)
Article
Physics, Fluids & Plasmas
Tiago P. Peixoto
Summary: We propose a method for inferring community structure in directed networks with an ordered hierarchical structure, utilizing a modified stochastic block model. Our approach combines ordered hierarchies with arbitrary mixing patterns between groups and includes directed degree correction to distinguish between nonlocal hierarchical structure and local in-and-out-degree imbalance. We demonstrate the application of our method on various empirical networks in different domains.
Article
Computer Science, Artificial Intelligence
Enrique G. Rodrigo, Juan C. Alfaro, Juan A. Aledo, Jose A. Gamez
Summary: This article proposes two alternative methods to improve the label ranking trees (LRTs) algorithm. These methods use distance-based criteria to select the best split at each node and can handle incomplete rankings efficiently. Experimental results show that the proposed methods are significantly faster and at least as accurate as the original Mallows-based LRT algorithm.
Article
Automation & Control Systems
Amit Praseed, Jelwin Rodrigues, P. Santhi Thilagam
Summary: In the past few decades, the growth of social networking sites has led to an unprecedented level of information distribution. Fake news detection in regional languages, such as Hindi, has posed challenges due to limited resources and the need for translation. Pre-trained transformer models, like BERT, ELECTRA, and RoBERTa, have shown promise in detecting fake news in multiple languages. This study proposes a method that uses an ensemble of pre-trained transformer models, XLM-RoBERTa, mBERT, and ELECTRA, to more efficiently detect fake news in resource-constrained languages like Hindi.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Konstantinos Nikolaidis, Thomas Plagemann, Stein Kristiansen, Vera Goebel, Mohan Kankanhalli
Summary: Improper or erroneous labelling can hinder reliable generalization for supervised learning, especially in critical fields like healthcare. This study presents an effective approach for learning under extreme label noise in medical applications, utilizing under-trained deep ensembles to improve generalization. Performance improvement from 0.02 to 0.55 was observed in sleep apnea detection tasks.
Article
Physics, Fluids & Plasmas
Anjuman Ara Khatun, Haider Hasan Jafri, Nirmal Punetha
Summary: This study demonstrates the control of chimera states in coupled chaotic oscillator networks through linear augmentation technique, altering the size and spatial location of different phase states. The effects of linear augmentation on multistable behavior and chimera states were analyzed through basin of attraction and stability analysis, showcasing the applicability of the technique to various types of oscillator networks. This research suggests that linear augmentation control can effectively manipulate chimera states and achieve desired collective dynamics in ensembles.
Article
Astronomy & Astrophysics
N. Crescini, G. Carugno, P. Falferi, A. Ortolan, G. Ruoso, C. C. Speake
Summary: Spin-dependent fifth forces, associated with particles beyond the standard model, can be studied by varying the distance between mass and spin and detecting the interactions using precision magnetometry techniques. Our measurement places stringent constraints on these interactions and demonstrates the potential for further exploration of the fifth force's parameter space.
Article
Multidisciplinary Sciences
Denis Abu Sammour, James L. Cairns, Tobias Boskamp, Christian Marsching, Tobias Kessler, Carina Ramallo Guevara, Verena Panitz, Ahmed Sadik, Jonas Cordes, Stefan Schmidt, Shad A. Mohammed, Miriam F. Rittel, Mirco Friedrich, Michael Platten, Ivo Wolf, Andreas von Deimling, Christiane A. Opitz, Wolfgang Wick, Carsten Hopf
Summary: Mass spectrometry imaging utilizes traditional ion images for metabolite visualization and analysis, but lacks consideration for nonlinearities and statistical significance. The computational framework moleculaR aims to improve signal reliability by Gaussian-weighting ion intensities and introduces probabilistic molecular mapping for statistically significant spatial abundance analysis. It allows cross-tissue comparisons and spatial statistical significance evaluation, facilitating the investigation of ion milieus, lipid remodeling pathways, and complex scores within the same image.
NATURE COMMUNICATIONS
(2023)
Article
Computer Science, Information Systems
Sevvandi Kandanaarachchi
Summary: Ensemble learning combines multiple algorithms or models to improve predictive performance and is used in various fields. This study introduces the use of Item Response Theory (IRT), a model used in educational psychometrics, to construct an unsupervised anomaly detection ensemble. The effectiveness of the IRT ensemble is demonstrated and shown to outperform other ensemble techniques, even with low correlation values among the anomaly detection methods.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Information Systems
Akanksha Mukhriya, Rajeev Kumar
Summary: We discuss the issue of score combinations in outlier detection ensembles (ODEs). Despite normalization, ODE score combinations may still be biased. Determining suitable normalization and avoiding dominance of specific detectors is challenging. We propose a framework called FairComb to address this issue and promote fairer combinations in ODEs.
INFORMATION SCIENCES
(2023)
Article
Agronomy
Aksana Yarashynskaya, Piotr Prus
Summary: This paper addresses the slow pace of Precision Agriculture (PA) introduction in developed, transitioning, and developing countries, focusing on PA adoption in Poland as a case study. By identifying PA adoption factors and ranking the adoption potential of Polish voivodships, this study contributes to the literature on transitioning economies.
