检索结果-内蒙古大学图书馆

2022 MACLEAN: MAChine Learning for EArth ObservatioN Workshop, MACLEAN 2022

作者： Ashiotis, Giannis Tsigkanos, Evgenios Christoudias, Theodoros Nicolaou, Mihalis A. Computation-Based Science and Technology Research Center The Cyprus Institute Nicosia Cyprus Climate and Atmosphere Research Center The Cyprus Institute Nicosia Cyprus

Air pollution is detrimental to the environment and contributes to the global burden of disease. Accurate prediction of surface concentrations of atmospheric pollutants, including fine particulate matter (PM2.5) and gaseous N O2, is key to mitigating these harmful effects. However, most forecasts produced by state-of-the-art models exhibit biases and limitations due to low-resolution (typically ∼ 40 km) and intense computational requirements. While deep learning-based approaches promise cheap inference, and can inherently scale and assimilate large amounts of data, ground truth measurements are obtained by station networks that are costly and sparsely distributed spatially - making the training of data-hungry deep learning methods tedious. In this context, we propose a novel deep learning framework that can accurately downscale the resolution of surface concentrations down to 1 × 1 km by leveraging additional data sources of static and time-varying nature (such as elevation and land use type). To do so, we carefully design a tailored data alignment and sampling process that balances the distribution across land use type, enhancing model generalization. To handle ground truth sparsity, we propose a tailored loss function that leads the network to model correlations across data sources close to ground station, while encouraging it to approximate model-based outputs elsewhere. With a set of qualitative and quantitative experiments, we demonstrate that the proposed method outperforms all compared modeling- and learning-based approaches, in both the aerosol and gaseous phase, with a significant increase in accuracy (up to 19% for N O2 and 13% for PM2.5) - while generalizing the improvement across the various land use types. © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

关键词： Data fusion

来源：评论

学校读者我要写书评

暂无评论

Proton and neutron electromagnetic form factors using Nf=2+1+1 twisted-mass fermions with physical values of the quark masses

arXiv

引用

arXiv 2025年

作者： Alexandrou, Constantia Bacchio, Simone Koutsou, Giannis Prasad, Bhavna Spanoudes, Gregoris Department of Physics University of Cyprus Cyprus Computation-based Science and Technology Research Center The Cyprus Institute Cyprus

We compute the electromagnetic form factors of the proton and neutron using lattice QCD with Nf = 2 + 1 + 1 twisted mass clover-improved fermions and quark masses tuned to their physical values. Three ensembles with lattice spacings of a=0.080 fm, 0.068 fm, and 0.057 fm, and approximately the same physical volume allow us to obtain the continuum limit directly at the physical pion mass. Several values of the source-sink time separation ranging from 0.5 fm to 1.5 fm are used, enabling a thorough analysis of excited state effects via multi-state fits. The disconnected contributions are analyzed using high statistics for the two-point functions combined with low-mode deflation and hierarchical probing for the fermion loop estimation. We study the momentum dependence of the form factors using the z-expansion and dipole Ansaetze, thereby enabling the extraction of the electric and magnetic radii, as well as the magnetic moments in the continuum limit, for which we provide preliminary results. © 2025, CC BY-NC-ND.

关键词： Fermions

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Sinogram Denoising for Single Photon Emission Computed Tomography

Unsupervised Sinogram Denoising for Single Photon Emission C...

引用

IEEE Symposium on Nuclear science (NSS/MIC)

作者： C. Chrysostomou The Cyprus Institute Computation-Based Science and Technology Research Cente Nicosia Cyprus

This paper presents a novel methodology for denoising sinograms based on cycle-consistent adversarial networks, focusing on medical imaging applications. The proposed method is designed to denoise sinograms without requiring paired information of low to high levels of noise and uncertainty, which is often difficult to obtain in real-world scenarios. The method was trained on three two sets of projection data from randomly generated software phantoms, using GATE, with various levels of collections time. The denoised and high-collection-time sinograms were then reconstructed and compared using the Ordered Subset Expectation Maximization (OSEM) method, which showed that the proposed method significantly removes noise without impacting the quality of the sinograms and improving the reconstructions. Finally, the method was tested on dopamine transporter imaging with single-photon emission computed tomography (DAT-SPECT) sinograms to demonstrate its effectiveness and capability.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Advanced Sinogram Reconstruction via Fourier Neural Operators for High-Quality SPECT Imaging

Advanced Sinogram Reconstruction via Fourier Neural Operator...

引用

IEEE Symposium on Nuclear science (NSS/MIC)

作者： C. Chrysostomou Computation-Based Science and Technology Research Cente The Cyprus Institute Nicosia Cyprus

This study introduces an innovative sinogram reconstruction technique employing Fourier neural operators (FNOs) for single-photon emission computed tomography (SPECT) imaging. The proposed method accepts a limited number of sinogram projections and directly reconstructs a superior image without intermediate steps. The approach is trained on a dataset of high-quality SPECT images using FNOs, a category of neural networks capable of learning linear operators through Fourier analysis. The method is assessed using software-generated phantoms via GATE and compared against the standard ordered subsets expectation maximization (OSEM) reconstruction method. The results demonstrate that the proposed method considerably enhances reconstruction quality and surpasses state-of-the-art compressed sensing techniques. The presented method holds the potential to decrease radiation exposure and scanning time required for SPECT imaging, rendering it a valuable asset in clinical practice. Future research could investigate the application of this method to other imaging modalities and advanced neural network architectures.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Novel approach for computing gradients of physical observables

arXiv

引用

arXiv 2023年

作者： Bacchio, Simone Computation-based Science and Technology Research Center The Cyprus Institute Nicosia Cyprus

We show that an infinitesimal step of gradient flow can be used for defining a novel approach for computing gradients of physical observables with respect to action parameters. Compared to the commonly used perturbative expansion, this approach does not require calculating any disconnected contribution or vacuum expectation value and can provide results up to three orders of magnitudes more precise. On the other hand, it requires a non-trivial condition to be satisfied by the flow action, the calculation of its force and its Laplacian, and the force of the observable, whose gradient needs to be measured. As a proof of concept, we measure gradients in β of Wilson loops in a four-dimensional SU(3) Yang-Mills theory simulated on a 164 lattice using the Wilson action. © 2023, CC BY.

关键词： Lattice theory

来源：评论

学校读者我要写书评

暂无评论

Efficient Learning of Multiple NLP Tasks via CollectiveWeight Factorization on BERT

Efficient Learning of Multiple NLP Tasks via CollectiveWeigh...

引用

2022 Findings of the Association for computational Linguistics: NAACL 2022

作者： Papadopoulos, Christos C. Panagakis, Yannis Koubarakis, Manolis Nicolaou, Mihalis A. Dept. of Informatics and Telecommunications National and Kapodistrian University of Athens Greece Computation-based Science and Technology Research Center The Cyprus Institute Cyprus

ISBN: (纸本)9781955917766

The Transformer architecture continues to show remarkable performance gains in many Natural Language Processing tasks. However, obtaining such state-of-the-art performance in different tasks requires fine-tuning the same model separately for each task. Clearly, such an approach is demanding in terms of both memory requirements and computing power. In this paper, aiming to improve training efficiency across multiple tasks, we propose to collectively factorize the weighs of the multi-head attention module of a pre-trained Transformer. We test our proposed method on finetuning multiple natural language understanding tasks by employing BERT-Large as an instantiation of the Transformer and the GLUE as the evaluation benchmark. Experimental results show that our method requires training and storing only 1% of the initial model parameters for each task and matches or improves the original fine-tuned model's performance for each task while effectively decreasing the parameter requirements by two orders of magnitude. Furthermore, compared to well-known adapter-based alternatives on the GLUE benchmark, our method consistently reaches the same levels of performance while requiring approximately four times fewer total and trainable parameters per task. © Findings of the Association for computational Linguistics: NAACL 2022 - Findings.

关键词： Natural language processing systems

来源：评论

学校读者我要写书评

暂无评论

DERIVING COARSE-GRAINED MODELS OF MOLECULAR SYSTEMS BY APPROXIMATING THE FREE ENERGY SURFACE WITH MACHINE LEARNING ALGORITHMS 9

DERIVING COARSE-GRAINED MODELS OF MOLECULAR SYSTEMS BY APPRO...

引用

9th ECCOMAS Thematic Conference on computational Methods in Structural Dynamics and Earthquake Engineering, COMPDYN 2023

作者： Bakas, Nikolaos P. Chazirakis, Antonis Christofi, Eleftherios Harmandaris, Vangelis Computation-based Science and Technology Research Center The Cyprus Institute 20 Konstantinou Kavafi Street Aglantzia Nicosia2121 Cyprus Department of Applied Mathematics University of Crete Heraklion CreteGR-71409 Greece

Atomistic molecular simulations are capable to predict the properties of systems, and materials, starting from their atomic microstructure. To enhance the range of spatiotemporal scales accessible by simulations coarse-grained (CG) models, that reduce the dimensionality of the underlying system, are used. In such models CG particles, representing group of atoms, are introduced. The derivation of accurate effective interactions between the CG particles (CG force field) is a grand challenge in order to study realistic complex systems associatted with important technological applications. The problem of coarse-grained simplification of the representation of the underlying atomistic microstructure (or molecular potential) is a hard task in the generic case. Significant information, such as the orientation of atoms, might be lost when using the Cartesian coordinates of the participating molecules to develop a model predicting the potential. Furthermore, the actual relationships among the input coordinates and the target potential stem from vastly complex interactions among the neighbouring molecules with the central one, as well as the corresponding atoms. Accordingly, even if we use vastly big datasets and supercomputing for training the models, it is not known whether an approximation algorithm yielding low accuracy is a weak learner or important features have been committed during the coarse-grained representation. In this work, we provide a detailed investigation of several algorithms that are used to approximate accurately the effective CG interaction (free energy surface) and present empirical evidence for their efficiency applied in a pool of databases for various materials. Particularly, we investigated tree-based models (gradient boosting and random forests), artificial neural networks' architectures, and higher-order polynomial regression with heuristic feature selection. The CG force field is automatically obtained from the differentiation of the ML model trai

关键词： Molecular dynamics

来源：评论

学校读者我要写书评

暂无评论

Thermal Tomography for Medical Applications

Thermal Tomography for Medical Applications

引用

IEEE Symposium on Nuclear science (NSS/MIC)

作者： C. Papanicolas A. Frixou L. Koutsantonis Computation-Based Science and Technology Research Center (CaSToRC) The Cyprus Institute Nicosia Cyprus

We report results supporting that Thermal Tomography may be feasible as a medical diagnostic tool. Tomographic images from investigations using software and hardware phantoms and the Athens Model Analysis Independent Scheme (AMIAS) analysis package show that it is feasible to detect in tomographic images low temperature (1-4 °C) hotspots at depths reaching 1,5 cm below the skin. These results indicate that identification and description of hotspots in this temperature range within the human body is possible, thereby allowing for the development of diagnostic capabilities for a number of diseases, such as rheumatoid arthritis and cancer where the tumor is situated in proximity to the skin surface.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Physics-Informed Deep Learning Approach for Reintroducing Atomic Detail in Coarse-Grained Configurations of Multiple Poly(lactic acid) Stereoisomers (vol 64, pg 1853, 2024)

引用

JOURNAL OF CHEMICAL INFORMATION AND MODELING 2024年第17期64卷 6926-6926页

作者： Christofi, Eleftherios Bacova, Petra Harmandaris, Vagelis A. Computation-based Science and Technology Research Center The Cyprus Institute Nicosia 2121 Cyprus Departamento de Ciencia de los Materiales e Ingeniería Metalúrgica y Química Inorgánica Facultad de Ciencias IMEYMAT Campus Universitario Río San Pedro s/n. Puerto Real Cádiz 11510 Spain Department of Mathematics and Applied Mathematics University of Crete Heraklion GR-71110 Greece Institute of Applied and Computational Mathematics Foundation for Research and Technology - Hellas Heraklion GR-71110 Crete Greece

Multiscale modeling of complex molecular systems, such as macromolecules, encompasses methods that combine information from fine and coarse representations of molecules to capture material properties over a wide range of spatiotemporal scales. Being able to exchange information between different levels of resolution is essential for the effective transfer of this information. The inverse problem of reintroducing atomistic degrees of freedom in coarse-grained (CG) molecular configurations is particularly challenging as, from a mathematical point of view, it is an ill-posed problem; the forward mapping from the atomistic to the CG description is typically defined via a deterministic operator (“one-to-one” problem), whereas the reversed mapping from the CG to the atomistic model refers to creating one representative configuration out of many possible ones (“one-to-many” problem). Most of the backmapping methods proposed so far balance accuracy, efficiency, and general applicability. This is particularly important for macromolecular systems with different types of isomers, i.e., molecules that have the same molecular formula and sequence of bonded atoms (constitution) but differ in the three-dimensional configurations of their atoms in space. Here, we introduce a versatile deep learning approach for backmapping multicomponent CG macromolecules with chiral centers, trained to learn structural correlations between polymer configurations at the atomistic level and their corresponding CG descriptions. This method is intended to be simple and flexible while presenting a generic solution for resolution transformation. In addition, the method is aimed to respect the structural features of the molecule, such as local packing, capturing therefore the physical properties of the material. As an illustrative example, we apply the model on linear poly(lactic acid) (PLA) in melt, which is one of the most popular biodegradable polymers. The framework is tested on a number of model sys

关键词： Biopolymers Copolymers Monomers Organic polymers Plastics

来源：评论

学校读者我要写书评

暂无评论

Condensation dynamics on inclined heterogeneous substrates

引用

International Journal of Thermofluids 2024年 23卷

作者： Vrionis, P.-Y. Demou, Andreas D. Savva, Nikos Computation-based Science and Technology Research Center The Cyprus Institute Aglantzia Nicosia2121 Cyprus Department of Mathematics and Statistics University of Cyprus Aglantzia Nicosia2109 Cyprus

This study presents numerical simulations of dropwise condensation on inclined chemically heterogeneous substrates in pure vapor atmosphere. To capture the dynamics of this process, a one-sided condensation model with a disjoining pressure is employed that incorporates the effects of gravity and chemical heterogeneities. We consider surfaces decorated with an array of square patches which are more hydrophilic compared to the rest of the surface, which act as nucleation, inception regions over which droplets form and grow and ultimately depin once their size becomes sufficiently large. The impact of the size and of the inception regions and the angle of inclination on the dynamics is considered, revealing significant qualitative and quantitative differences between the cases. For cases with purely structured substrates, it is observed that, as the number of inception regions increases, the condensation process transitions from steady to unsteady, with the critical number of inception regions increasing with the inclination angle increases. Further insight is provided by analyzing the snapshots of the liquid phase on the substrate, revealing the formation of liquid streaks or films, which influence both the dynamics of the flow and the total liquid volume on the substrate. The inclusion of random impurities on the structured substrate dramatically impacts the condensation dynamics, which is found to hinder the formation of liquid film, leading to a decrease in the total liquid volume generated on the substrate. © 2024 The Authors

关键词： Condensation

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：