检索结果-内蒙古大学图书馆

Robust Model Aggregation for Heterogeneous Federated Learning: Analysis and Optimizations

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Shao, Yumeng Li, Jun Shi, Long Wei, Kang Ding, Ming Li, Qianmu Li, Zengxiang Chen, Wen Jin, Shi The School of Electronic and Optical Engineering Nanjing University of Science and Technology Nanjing210094 China Data61 CSIRO Sydney Australia The Digital Economy Research Institute Nanjing University of Science and Technology Jiangsu Nanjing210094 China The Institute of High Performance Computing A*STAR Singapore The Digital Research Institute ENN Group China The Department of Electronic Engineering Shanghai Jiao Tong University Shanghai200240 China The National Mobile Communications Research Laboratory Southeast University Nanjing210096 China

Conventional synchronous federated learning (SFL) frameworks suffer from performance degradation in heterogeneous systems due to imbalanced local data size and diverse computing power on the client side. To address this problem, asynchronous FL (AFL) and semi-asynchronous FL have been proposed to recover the performance loss by allowing asynchronous aggregation. However, asynchronous aggregation incurs a new problem of inconsistency between local updates and global updates. Motivated by the issues of conventional SFL and AFL, we first propose a time-driven SFL (T-SFL) framework for heterogeneous systems. The core idea of T-SFL is that the server aggregates the models from different clients, each with varying numbers of iterations, at regular time intervals. To evaluate the learning performance of T-SFL, we provide an upper bound on the global loss function. Further, we optimize the aggregation weights to minimize the developed upper bound. Then, we develop a discriminative model selection (DMS) algorithm that removes local models from clients whose number of iterations falls below a predetermined threshold. In particular, this algorithm ensures that each client’s aggregation weight accurately reflects its true contribution to the global model update, thereby improving the efficiency and robustness of the system. To validate the effectiveness of T-SFL with the DMS algorithm, we conduct extensive experiments using several popular datasets including MNIST, Cifar-10, Fashion-MNIST, and SVHN. The experimental results demonstrate that T-SFL with the DMS algorithm can reduce the latency of conventional SFL by 50%, while achieving an average 3% improvement in learning accuracy over state-of-the-art AFL algorithms. Copyright © 2024, The Authors. All rights reserved.

关键词： Learning algorithms

1d Selective Confinement and Diffusion of Metal Atoms on Graphene

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Stavrić, Srdjan Chesnyak, Valeria del Puppo, Simone Panighel, Mirco Comelli, Giovanni Africh, Cristina Šljivančanin, Željko Peressi, Maria Consiglio Nazionale delle Ricerche CNR-SPIN c/o Università degli Studi "G. D’Annunzio" Chieti66100 Italy Vinča Institute of Nuclear Sciences National Institute of the Republic of Serbia University of Belgrade P. O. Box 522 BelgradeRS-11001 Serbia Physics Department University of Trieste via A. Valerio 2 Trieste34127 Italy CNR-IOM Laboratorio TASC S.S. 14 Km 163.5 Basovizza Trieste34149 Italy ICSC - Italian Research Center on High Performance Computing Big Data and Quantum Computing Italy

Graphene (G) grown on Ni(100) forms a striped moiré pattern of valleys, where G approaches the nickel substrate and interacts with it rather strongly, and ridges, where G stays far away from the substrate and acts almost free-standing. Combining density functional theory (DFT) calculations and scanning-tunneling microscopy (STM) measurements, we show that this peculiar moiré constitutes a regular nanostructured template on a 2D support, confining in 1D trails single metal atoms and few atoms clusters. DFT calculations show that the confinement is selective and highly dependent on the atomic specie, with some species preferring to adsorb on ridges and the other showing preference for valleys. The origin of such disparate behavior is traced back to the electrostatic interaction between the charged adsorbate and the nickel surface. Moreover, the selectivity is not restricted to the adsorption process only, but persists as adsorbate starts its diffusion, resulting in unidirectional mass transport on a continuous 2D support. © 2023, The Authors. All rights reserved.

关键词： Graphene

Supermassive black hole spin evolution in cosmological simulations with OpenGadget3

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Sala, Luca Valentini, Milena Biffi, Veronica Dolag, Klaus Universitäts-Sternwarte Fakultät für Physik Ludwig-Maximilians-Universität München Scheinerstr. 1 MünchenD-81679 Germany Astronomy Unit Department of Physics University of Trieste via Tiepolo 11 TriesteI-34131 Italy INAF – Osservatorio Astronomico di Trieste via Tiepolo 11 TriesteI-34131 Italy ICSC - Italian Research Center on High Performance Computing Big Data and Quantum Computing Italy Max-Planck-Institut für Astrophysik Karl-Schwarzschild-Str. 1 GarchingD-85741 Germany

Context. Mass and spin of massive black holes (BHs) at the centre of galaxies evolve due to gas accretion and mergers with other BHs. Besides affecting e.g. the evolution of relativistic jets, the BH spin determines the efficiency with which the BH radiates energy. Aims. Using cosmological, hydrodynamical simulations, we investigate the evolution of the BH spin across cosmic time and its role in controlling the joint growth of supermassive BHs and their host galaxies. Methods. We implement a sub-resolution prescription that models the BH spin, accounting for both BH coalescence and misaligned accretion through a geometrically thin, optically thick disc. We investigate how BH spin evolves in two idealised setups, in zoomed-in simulations, and in a cosmological volume. The latter simulation allows us to retrieve statistically robust results as for the evolution and distribution of BH spins as a function of BH properties. Results. We find that BHs with MBH 2 × 107 M☉ grow through gas accretion, occurring mostly in a coherent fashion that favours spin-up. Above MBH 2 × 107 M☉ the gas angular momentum directions of subsequent accretion episodes are often uncorrelated with each other. The probability of counter-rotating accretion and hence spin-down increases with BH mass. In the latter mass regime, BH coalescence plays an important role. The spin magnitude displays a wide variety of histories, depending on the dynamical state of the gas feeding the BH and the relative contribution of mergers and gas accretion. As a result of their combined effect, we observe a broad range of values of the spin magnitude at the high-mass end. Reorientation of the BH spin direction occurs on short timescales ( 10 Myr) only during highly-accreting phases (fEdd 0.1). Our predictions for the distributions of BH spin and spin-dependent radiative efficiency as a function of BH mass are in very good agreement with observations. © 2023, CC BY.

关键词： Numerical methods

From few to many maps: A fast map-level emulator for extreme augmentation of CMB systematics datasets

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Campeti, P. Delouis, J.-M. Pagano, L. Allys, E. Lattanzi, M. Gerbino, M. INFN Sezione di Ferrara Via Saragat 1 Ferrara44122 Italy ICSC Centro Nazionale "High Performance Computing Big Data and Quantum Computing" Univ. Brest CNRS Ifremer IRD Brest29200 France Dipartimento di Fisica e Scienze della Terra Università degli Studi di Ferrara via Saragat 1 FerraraI-44122 Italy Institut d’Astrophysique Spatiale CNRS Univ. Paris-Sud Université Paris-Saclay Bât. 121 Orsay91405 cedex France Laboratoire de Physique de l’Ecole Normale Supérieure ENS Univ. PSL CNRS Sorbonne Univ. Univ. Paris Cité Paris75005 France

We introduce a novel, fast, and efficient generative model built upon scattering covariances, the most recent iteration of the scattering transforms statistics. This model is designed to augment by several orders of magnitude the number of map simulations in datasets of computationally expensive Cosmic Microwave Background (CMB) instrumental systematics simulations, including their non-Gaussian and inhomogeneous features. Unlike conventional neural network-based algorithms, this generative model requires only a minimal number of training samples, making it highly compatible with the computational constraints of typical CMB simulation campaigns. While our primary focus is on spherical data, the framework is inherently versatile and readily applicable to 1D and 2D planar data, leveraging the localized nature of scattering statistics. We validate the method using realistic simulations of CMB systematics, which are particularly challenging to emulate, and perform extensive statistical tests to confirm its ability to produce new statistically independent approximate realizations. Remarkably, even when trained on as few as 10 simulations, the emulator closely reproduces key summary statistics—including the angular power spectrum, scattering coefficients, and Minkowski functionals—and provides pixel covariance estimates with substantially reduced sample noise compared to those obtained without augmentation. The proposed approach has the potential to shift the paradigm in simulation campaign design. Instead of producing large numbers of low- or medium-accuracy simulations, future pipelines can focus on generating a few high-accuracy simulations that are then efficiently augmented using such generative model. This promises significant benefits not only for current and forthcoming cosmological surveys such as Planck, LiteBIRD, Simons Observatory, CMB-S4, Euclid and Rubin-LSST, but also for diverse fields including oceanography and climate science. We make both the general fra

关键词： Cosmology

Classifying spectra of emission-line regions with neural networks An application to integral field spectroscopic data of M33

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Bracci, Caterina Belfiore, Francesco Ginolfi, Michele Feltre, Anna Mannucci, Filippo Marconi, Alessandro Cresci, Giovanni Bertola, Elena Bombini, Alessandro Ceci, Matteo Marconcini, Cosimo Moreschini, Bianca Scialpi, Martina Tozzi, Giulia Ulivi, Lorenzo Venturi, Giacomo Dipartimento di Fisica e Astronomia Università di Firenze Via G. Sansone 1 Sesto F.no FirenzeI-50019 Italy INAF - Osservatorio Astrofisico di Arcetri Largo E. Fermi 5 FlorenceI-50125 Italy Via Bruno Rossi 1 FI Sesto Fiorentino50019 Italy ICSC - Centro Nazionale di Ricerca in High Performance Computing Big Data & Quantum Computing Via Magnanelli 2 BO Casalecchio di Reno40033 Italy University of Trento Via Sommarive 14 TrentoI-38123 Italy Gießenbachstraße 1 Garching85748 Germany Scuola Normale Superiore Piazza dei Cavalieri 7 PisaI-56126 Italy

Emission-line regions are key to understanding the properties and evolution of galaxies, as they trace the exchange of matter and energy between stars and the interstellar medium (ISM). In nearby galaxies, individual nebulae can be resolved from the background, and classified through their optical spectral properties. Traditionally, the classification of these emission-line regions in H ii regions, planetary nebulae (PNe), supernova remnants (SNR), and diffuse ionised gas (DIG) relies on criteria based on single or multiple emission-line ratios. However, these methods face limitations due to the rigidity of the classification boundaries, the narrow scope of information they are based upon, and the inability to take line-of-sight superpositions of emission-line regions into account. In this work, we explore the use of artificial neural networks to classify emission-line regions using their optical spectra. Our training set consists of simulated optical spectra, obtained from photoionisation and shock models, and processed to match observations obtained with the MUSE integral field spectrograph at the ESO/VLT. We evaluate the performance of the network on simulated spectra exploring a range of signal-to-noise (S/N) levels, different values for dust extinction, and the superposition of different nebulae along the line of sight. At infinite S/N the network achieves perfect predictive performance, while, as the S/N decreases, the classification accuracy declines, reaching an average of ∼ 80% at S/N(Hα)=20. We then apply our model to real spectra from MUSE observations of the Local Group galaxy M33. The network provides a robust classification of individual spaxels, even at low S/N, identifying H ii regions and PNe and distinguishing them from SNRs and diffuse ionized gas, while identifying overlapping nebulae. Moreover, we compare the network’s classification with traditional diagnostics, finding a satisfactory level of agreement between the two approaches. We identify t

关键词： Supernovae

Systematic bias in dark siren statistical methods and its impact on Hubble constant measurement

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Alfradique, Viviane Bom, Clécio R. Castro, Tiago Centro Brasileiro de Pesquisas Físicas Rua Dr. Xavier Sigaud 150 RJ Rio de Janeiro22290-180 Brazil Centro Federal de Educação Tecnológica Celso Suckow da Fonseca Rodovia Márcio Covas lote J2 quadra J Itaguaí Brazil INAF Osservatorio Astronomico di Trieste via Tiepolo 11 TriesteI-34131 Italy INFN Sezione di Trieste TriesteI-34100 Italy IFPU Institute for Fundamental Physics of the Universe via Beirut 2 Trieste34151 Italy ICSC - Centro Nazionale di Ricerca in High Performance Computing Big Data e Quantum Computing Via Magnanelli 2 Bologna Italy

The advent of the multimessenger cosmology marked by the detection of GW170817, gravitational waves (GWs) from compact objects at cosmological distances demonstrated Standard Sirens as a relevant cosmological probe. In the absence of an electromagnetic counterpart identification, GWs carry valuable information through the dark siren approach, where the source redshift is estimated using galaxy catalogs of potential hosts within the localization volume. However, the dark siren analysis can be affected by galaxy catalog incompleteness at the limits of gravitational-wave detectability, potentially introducing biases in the constraints on cosmological parameters. Focusing on GWs from binary black holes (BBH) detected by the LIGO, Virgo, and KAGRA (LVK) collaboration, we explore the possible systematic biases in the measurement of the Hubble constant (H0). These biases may arise from (1) the incompleteness of catalogs due to the apparent magnitude thresholds of optical telescope sensitivity, and (2) the use of incorrect weighting schemes (e.g., using star formation or stellar mass as tracers of the host galaxy) for each potential host. We found that an unbiased estimate of H0 can be obtained when the corrected weighting scheme is applied to a complete or volume-limited catalog. The results using stellar mass as a tracer indicate that a percent-level measurement of H0, with a precision of approximately 3%, can be achieved with 100 binary black hole detections by the LVK at O4 and O5 sensitivity. The O5 run provides a slight improvement, reducing the H0 uncertainty by 0.07 km/s/Mpc compared to the O4-like configuration. This precision is achievable by using a complete galaxy catalog that covers the 90% of the localisation probability (A90%) for each GW detection, with A90% 2. The H0 precision increases to approximately 6% when it is assumed that every galaxy has an equal probability of being the host. © 2025, CC BY.

关键词： Cosmology

Improved Weak Lensing Photometric Redshift Calibration via StratLearn and Hierarchical Modeling

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Autenrieth, Maximilian Wright, Angus H. Trotta, Roberto van Dyk, David A. Stenning, David C. Joachimi, Benjamin Department of Mathematics Imperial College London 180 Queen’s Gate LondonSW7 2AZ United Kingdom German Centre for Cosmological Lensing Bochum44780 Germany SISSA International School for Advanced Studies Via Bonomea 265 Trieste34136 Italy Department of Physics Imperial College London Blackett Laboratory Prince Consort Rd LondonSW72AZ United Kingdom Italian Research Center on High Performance Computing Big Data and Quantum Computing Italy INFN – National Institute for Nuclear Physics Via Valerio 2 Trieste34127 Italy Department of Statistics and Actuarial Science Simon Fraser University Canada Department of Physics and Astronomy University College London Gower Street LondonWC1E 6BT United Kingdom

Discrepancies between cosmological parameter estimates from cosmic shear surveys and from recent Planck cosmic microwave background measurements challenge the ability of the highly successful ΛCDM model to describe the nature of the Universe. To rule out systematic biases in cosmic shear survey analyses, accurate redshift calibration within tomographic bins is key. In this paper, we improve photo-z calibration via Bayesian hierarchical modeling of full galaxy photo-z conditional densities, by employing StratLearn, a recently developed statistical methodology, which accounts for systematic differences in the distribution of the spectroscopic training/source set and the photometric target set. Using realistic simulations that were designed to resemble the KiDS+VIKING-450 dataset, we show that StratLearn-estimated conditional densities improve the galaxy tomographic bin assignment, and that our StratLearn-Bayesian framework leads to nearly unbiased estimates of the target population means. This leads to a factor of ∼ 2 improvement upon the previously best photo-z calibration method. Our approach delivers a maximum bias per tomographic bin of ∆〈z〉 = 0.0095 ± 0.0089, with an average absolute bias of 0.0052 ± 0.0067 across the five tomographic bins. © 2024, CC BY.

关键词： Galaxies

Bayesian evidence estimation from posterior samples with normalizing flows

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Srinivasan, Rahul Crisostomi, Marco Trotta, Roberto Barausse, Enrico Breschi, Matteo SISSA Via Bonomea 265 Trieste34136 Italy INFN Sezione di Trieste Trieste34149 Italy IFPU - Institute for Fundamental Physics of the Universe Via Beirut 2 Trieste34014 Italy TAPIR Division of Physics Mathematics and Astronomy California Institute of Technology PasadenaCA91125 United States Dipartimento di Fisica Università di Pisa Largo Bruno Pontecorvo 3 Pisa56127 Italy Department of Physics Imperial College London Blackett Lab Prince Consort Rd LondonSW7 2AZ United Kingdom Italian Research Center on High Performance Computing Big Data and Quantum Computing 13 Via Magnanelli 2 40033 Italy

We propose a novel method (floZ), based on normalizing flows, to estimate the Bayesian evidence (and its numerical uncertainty) from a pre-existing set of samples drawn from the unnormalized posterior distribution. We validate it on distributions whose evidence is known analytically, up to 15 parameter space dimensions, and compare with two state-of-the-art techniques for estimating the evidence: nested sampling (which computes the evidence as its main target) and a k-nearest-neighbors technique that produces evidence estimates from posterior samples. Provided representative samples from the target posterior are available, our method is more robust to posterior distributions with sharp features, especially in higher dimensions. For a simple multivariate Gaussian, we demonstrate its accuracy for up to 200 dimensions with 105 posterior samples. floZ has wide applicability, e.g., to estimate evidence from variational inference, Markov Chain Monte Carlo samples, or any other method that delivers samples and their likelihood from the unnormalized posterior density. As a physical application, we use floZ to compute the Bayes factor for the presence of the first overtone in the ringdown signal of the gravitational wave data of GW150914, finding good agreement with nested sampling. Copyright © 2024, The Authors. All rights reserved.

关键词： Markov chains

SIDE-real: Supernova Ia Dust Extinction with truncated marginal neural ratio estimation applied to real data

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Karchev, Konstantin Grayling, Matthew Boyd, Benjamin M. Trotta, Roberto Mandel, Kaisey S. Weniger, Christoph Via Bonomea 265 TriesteI-34136 Italy Institute of Astronomy Kavli Institute for Cosmology Madingley Road CambridgeCB3 0HA United Kingdom Astrophysics Group Physics Department Blackett Lab Imperial College London Prince Consort Road LondonSW7 2AZ United Kingdom INFN - National Institute for Nuclear Physics Via Valerio 2 TriesteI-34127 Italy Italian Research Center on High-Performance Computing Big Data and Quantum Computing Via Magnanelli 2 BO Casalecchio di RenoI-40033 Italy Statistical Laboratory DPMMS University of Cambridge Wilberforce Road CambridgeCB3 0WB United Kingdom University of Amsterdam Science Park 904 AmsterdamNL-1098 XH Netherlands

We present the first fully simulation-based hierarchical analysis of the light curves of a population of low-redshift type Ia supernovæ (SNæ Ia). Our hardware-accelerated forward model, released in the Python package slicsim, includes stochastic variations of each SN's spectral flux distribution (based on the pre-trained BayeSN model), extinction from dust in the host and in the Milky Way, redshift, and realistic instrumental noise. By utilizing truncated marginal neural ratio estimation (TMNRE), a neural network-enabled simulation-based inference technique, we implicitly marginalize over 4000 latent variables (for a set of ≈100 SNæ Ia) to efficiently infer SN Ia absolute magnitudes and host-galaxy dust properties at the population level while also constraining the parameters of individual objects. Amortization of the inference procedure allows us to obtain coverage guarantees for our results through Bayesian validation and frequentist calibration. Furthermore, we show a detailed comparison to full likelihood-based inference, implemented through Hamiltonian Monte Carlo, on simulated data and then apply TMNRE to the light curves of 86 SNæ Ia from the Carnegie Supernova Project, deriving marginal posteriors in excellent agreement with previous work. Given its ability to accommodate arbitrarily complex extensions to the forward model, e.g. different populations based on host properties, redshift evolution, complicated photometric redshift estimates, selection effects, and non-Ia contamination, without significant modifications to the inference procedure, TMNRE has the potential to become the tool of choice for cosmological parameter inference from future, large SN Ia samples. © 2024, CC BY.

关键词： Supernovae