检索结果-内蒙古大学图书馆

Proceedings of the 38th International Conference on Neural Information Processing Systems

作者： Weilin Lin Li Liu Shaokui Wei Jianze Li Hui Xiong The Hong Kong University of Science and Technology (Guangzhou) The Chinese University of Hong Kong Shenzhen Shenzhen International Center for Industrial and Applied Mathematics and Shenzhen Research Institute of Big Data and The Chinese University of Hong Kong Shenzhen

ISBN: (纸本)9798331314385

The security threat of backdoor attacks is a central concern for deep neural networks (DNNs). Recently, without poisoned data, unlearning models with clean data and then learning a pruning mask have contributed to backdoor defense. Additionally, vanilla fine-tuning with those clean data can help recover the lost clean accuracy. However, the behavior of clean unlearning is still under-explored, and vanilla fine-tuning unintentionally induces back the backdoor effect. In this work, we first investigate model unlearning from the perspective of weight changes and gradient norms, and find two interesting observations in the backdoored model: 1) the weight changes between poison and clean unlearning are positively correlated, making it possible for us to identify the backdoored-related neurons without using poisoned data; 2) the neurons of the backdoored model are more active (i.e., larger gradient norm) than those in the clean model, suggesting the need to suppress the gradient norm during fine-tuning. Then, we propose an effective two-stage defense method. In the first stage, an efficient Neuron Weight Change (NWC)-based Backdoor Reinitialization is proposed based on observation 1). In the second stage, based on observation 2), we design an Activeness-Aware Fine-Tuning to replace the vanilla fine-tuning. Extensive experiments, involving eight backdoor attacks on three benchmark datasets, demonstrate the superior performance of our proposed method compared to recent state-of-the-art backdoor defense approaches. The code is available at https://***/linweiii/***.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A thorough investigation of the prospects of eLISA in addressing the Hubble tension: Fisher Forecast, MCMC and Machine Learning

arXiv

引用

arXiv 2023年

作者： Shah, Rahul Bhaumik, Arko Mukherjee, Purba Pal, Supratik Physics and Applied Mathematics Unit Indian Statistical Institute 203 B.T. Road Kolkata700 108 India Technology Innovation Hub on Data Science Big Data Analytics and Data Curation Indian Statistical Institute 203 B.T. Road Kolkata700 108 India

We carry out an in-depth analysis of the capability of the upcoming space-based gravitational wave mission eLISA in addressing the Hubble tension, with a primary focus on observations at intermediate redshifts (3 0) at the sub-percent level. MCMC/GP results predict reduced tensions for models/fiducials which are currently harder to reconcile with direct measurements of H0, whereas no significant change occurs for models/fiducials at lesser tensions with the latter. This feature warrants further investigation in this direction. Copyright © 2023, The Authors. All rights reserved.

关键词： Machine learning

来源：评论

学校读者我要写书评

暂无评论

Primordial magnetic non-Gaussianity with generic vacua and detection prospects in CMB spectral distortions

arXiv

引用

arXiv 2023年

作者： Bhaumik, Arko Pal, Supratik Physics and Applied Mathematics Unit Indian Statistical Institute 203 B.T. Road Kolkata700 108 India Technology Innovation Hub on Data Science Big Data Analytics and Data Curation Indian Statistical Institute 203 B.T. Road Kolkata700 108 India

Assuming a slow-roll inflationary model where conformal invariance of the Maxwell action is broken via a non-minimal kinetic coupling term, we investigate the non-Gaussian three-point cross-correlation function between the primordial curvature perturbation and the primordial magnetic field, under a fairly general choice of initial vacua for both the scalar and the gauge field sectors. Among the possible triangular configurations of the resulting cross-bispectrum, we find that the squeezed limit leads to local-type non-Gaussianity allowing a product form decomposition in terms of the scalar and magnetic power spectra, which is a generic result independent of any specific choice of the initial states. We subsequently explore its detection prospects in the CMB via correlations between pre-recombination µ-type spectral distortions and temperature anisotropies, sourced by such a primordial cross-correlation. Our analysis with several proposed next-generation CMB missions forecasts a low value of the signal-to-noise ratio (SNR) for the µT spectrum if both the vacua are assumed to be pure Bunch-Davies. On the contrary, the SNR may be enhanced significantly for non-Bunch-Davies initial states for the magnetic sector within allowed bounds from current CMB data. Copyright © 2023, The Authors. All rights reserved.

关键词： Signal to noise ratio

来源：评论

学校读者我要写书评

暂无评论

Inverse Evolution Layers: physics-informed Regularizers for Deep Neural Networks

arXiv

引用

arXiv 2023年

作者： Liu, Chaoyu Qiao, Zhonghua Li, Chao Schönlieb, Carola-Bibiane Department of Applied Mathematics and Theoretical Physics University of Cambridge United Kingdom Department of Applied Mathematics Research Institute for Smart Energy The Hong Kong Polytechnic University Hong Kong School of Science and Engineering School of Medicine University of Dundee United Kingdom

Traditional image processing methods employing partial differential equations (PDEs) offer a multitude of meaningful regularizers, along with valuable theoretical foundations for a wide range of image-related tasks. This makes their integration into neural networks a promising avenue. In this paper, we introduce a novel regularization approach inspired by the reverse process of PDE-based evolution models. Specifically, we propose inverse evolution layers (IELs), which serve as bad property amplifiers to penalize neural networks of which outputs have undesired characteristics. Using IELs, one can achieve specific regularization objectives and endow neural networks’ outputs with corresponding properties of the PDE models. Our experiments, focusing on semantic segmentation tasks using heat-diffusion IELs, demonstrate their effectiveness in mitigating noisy label effects. Additionally, we develop curve-motion IELs to enforce convex shape regularization in neural network-based segmentation models for preventing the generation of concave outputs. Theoretical analysis confirms the efficacy of IELs as an effective regularization mechanism, particularly in handling training with label issues. Copyright © 2023, The Authors. All rights reserved.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Turbulent Length Scale for Multilayer RANS Model of Urban Canopy and Its Evaluation Based on Large-Eddy Simulations

Supercomputing Frontiers and Innovations

引用

Supercomputing Frontiers and Innovations 2021年第4期8卷 100-116页

作者： Glazunov, Andrey V. Debolskiy, Andrey V. Mortikov, Evgeny V. G.I. Marchuk Institute of Numerical Mathematics Russian Academy of Science Moscow Russia Moscow Center for Fundamental and Applied Mathematics Moscow Russia Research Computing Center Lomonosov Moscow State University Moscow Russia A.M. Obukhov Institute of Atmospheric Physics Russian Academy of Science Moscow Russia

Large-Eddy Simulation (LES) numerical experiments of neutrally-stratified turbulent flow over an urban-type surface and passive scalar transport by this flow are performed. A simple parameterization of the turbulent length scale containing only one empirical constant is proposed. Multilayer Reynolds-Averaged Navier-Stokes (RANS) model of turbulent flow and turbulent scalar diffusion is constructed. The results of the RANS model are compared with the LES experiments. It is shown that the proposed approach allows predicting the average flow velocity and the scalar concentration inside and above the urban canopy. © The Authors 2021. This paper is published with open access at ***

关键词： Large eddy simulation

来源：评论

学校读者我要写书评

暂无评论

Stability Analysis and Controlling Chaos of Fractional-Order Three-Species Food Chain Model with Fear

引用

International Journal of applied and Computational mathematics 2021年第6期7卷 1-14页

作者： Mishra, Amit K. Kumar, Rajneesh Yadav, Vijay K. Department of Mathematics Shree M P Shah Arts Science College Surendranagar 363001 Gujarat India Department of Physics Institute of Science Banaras Hindu University Varanasi 221005 UP India Department of Applied Mathematics Amity School of Applied Sciences Amity University Gurugram 122413 Haryana India

In this article, stability analysis and chaos control of three species food chain model have been studied in the fractional order case. The fractional order three-species food chain model with fear term is the extension work of Panday et al. (Int J Bifurc Chaos 28:1850009, 2018), where they discussed the stability analysis and control of chaos of the model under several conditions. The stability analysis is done using Routh–Hurwitz condition and Matignon theorem for fractional order differential equations. The analysis has been done for the fractional-order of a three species food chain model with and without fear terms k₁ and k₂. If the fear terms are neglected in the considered model then the model is reduced to Hastings-Powell model. The chaos control is done using Lyapunov stability theory, the equilibrium points of the system are computed and stability is obtained in the system. The results are obtained through numerical simulation. © 2021, The Author(s), under exclusive licence to Springer Nature India Private Limited.

关键词： Chaos control Fractional derivative Stability analysis Three species food chain model

来源：评论

学校读者我要写书评

暂无评论

Reconstructing the Hubble parameter with future Gravitational Wave missions using Machine Learning

arXiv

引用

arXiv 2023年

作者： Mukherjee, Purba Shah, Rahul Bhaumik, Arko Pal, Supratik Physics and Applied Mathematics Unit Indian Statistical Institute 203 B.T. Road Kolkata700 108 India Technology Innovation Hub on Data Science Big Data Analytics and Data Curation Indian Statistical Institute 203 B.T. Road Kolkata700 108 India

We study the prospects of Gaussian processes (GP), a machine learning (ML) algorithm, as a tool to reconstruct the Hubble parameter H(z) with two upcoming gravitational wave missions, namely the evolved Laser Interferometer Space Antenna (eLISA) and the Einstein Telescope (ET). Assuming various background cosmological models, the Hubble parameter has been reconstructed in a non-parametric manner with the help of GP using realistically generated catalogs for each mission. The effects of early-time and late-time priors on the reconstruction of H(z), and hence on the Hubble constant (H0), have also been focused on separately. Our analysis reveals that GP is quite robust in reconstructing the expansion history of the Universe within the observational window of the specific missions under consideration. We further confirm that both eLISA and ET would be able to provide constraints on H(z) and H0 which would be competitive to those inferred from current datasets. In particular, we observe that an eLISA run of ∼ 10-year duration with ∼ 80 detected bright siren events would be able to constrain H0 as good as a ∼ 3-year ET run assuming ∼ 1000 bright siren event detections. Further improvement in precision is expected for longer eLISA mission durations such as a ∼ 15-year time-frame having ∼ 120 events. Lastly, we discuss the possible role of these future gravitational wave missions in addressing the Hubble tension, for each model, on a case-by-case basis. Copyright © 2023, The Authors. All rights reserved.

关键词： Gravity waves

来源：评论

学校读者我要写书评

暂无评论

AI for Photonics and Photonics for AI

AI for Photonics and Photonics for AI

引用

Optics and Photonics (ICOP), Italian Conference on

作者： Giulia Ciacci Cristiano D'Andrea Paolo Matteini Elisa Maria Alessi Sara Di Ruzza Maria Antonietta Pascali Sara Colantonio Ambra Giannetti Andrea Barucci Institute of Applied Physics “Nello Carrara” Italian National Research Council Sesto Fiorentino Italy Institute for Applied Mathematics and Information Technologies “Enrico Magenes” Italian National Research Council Milano Italy Department of Mathematics and Applications University of Palermo Palermo Italy Institute of Information Science and Technologies “Alessandro Faedo” Italian National Research Council Pisa Italy

ISBN: (数字)9798350361803

ISBN: (纸本)9798350361810

Artificial Intelligence (AI) is a disrupting technology, which has been changing our world. Photonics is a well-established technology, encompassing the generation, emission, transmission, modulation, signal processing, switching, amplification, and detection of light. The aim of this work is two-fold. From one side we aim to explore how AI methods can be used in Photonics applications, understanding the ubiquity of such models thanks to a journey through some authors' Machine Learning applications, going from images in Radiomics up to time series analysis. We will show strengths and weaknesses of AI in Photonics, along with new possible perspectives. In addition, we will explore the integration of Photonics and AI, which is a burgeoning field that leverages the strengths of both technologies to overcome limitations inherent in traditional electronic-based computing systems. Actually, AI models, especially those utilizing Artificial Neural Networks, require substantial computational power and speed, which can be enhanced through photonic technologies. This synergy between AI and Photonics not only can accelerate computational processes but also opens new avenues for implementing advanced Machine Learning models in real-time applications, showcasing a significant paradigm shift in the field of computing.

关键词： Protocols Computational modeling Time series analysis Signal processing algorithms Machine learning Artificial neural networks Real-time systems Artificial intelligence Photonics Radiomics

来源：评论

学校读者我要写书评

暂无评论

An Intelligent Hexahedral Mesh Smoothing Method Based on Deep Learning

SSRN

引用

SSRN 2023年

作者： Liu, Lufeng Zhou, Xuan Wang, Yutao Li, Shuixiang Institute of Applied Physics and Computational Mathematics Beijing100094 China Department of Mechanics and Engineering Science College of Engineering Peking University Beijing100871 China

We introduce an intelligent mesh smoothing method for the mesh quality improvement of unstructured 3D hexahedral meshes with fixed boundaries. We apply a supervised learning framework based on the deep neural networks. We directly generate the training data from the high-quality meshes and their stochastic disturbance. The supervised learning model will build a direct mapping from the poor-quality mesh to the high-quality mesh via the training data, making it possible to overcome the complicated design of traditional smoothing methods. Using only neighboring node coordinates and constructing them in point cloud format, the trained neural network can smooth both uniform and non-uniform meshes. The intelligent mesh smoothing method generates the smoothed mesh with better mesh quality and fewer iteration steps, compared to the Laplacian method and an optimization-based method. Moreover, the trained network demonstrates high generalization capability on the mesh with different disturbance forms which are not concerned in the training data. To the best of our knowledge, this is the first time that the machine learning method has been applied to the smoothing of 3D unstructured hexahedral meshes and demonstrates good application prospect. © 2023, The Authors. All rights reserved.

关键词： Supervised learning

来源：评论

学校读者我要写书评

暂无评论

Efficient modeling of sub-kilometer surface wind with Gaussian processes and neural networks

arXiv

引用

arXiv 2024年

作者： Zanetta, Francesco Nerini, Daniele Buzzi, Matteo Moss, Henry Institute for Atmospheric and Climate Science ETHZ Zürich Switzerland Federal Institute of Climatology and Meteorology MeteoSwiss Monti Locarno Switzerland Department of Applied Mathematics and Theoretical Physics University of Cambridge Cambridge United Kingdom

Accurately representing surface weather at the sub-kilometer scale is crucial for optimal decision-making in a wide range of applications. This motivates the use of statistical techniques to provide accurate and calibrated probabilistic predictions at a lower cost compared to numerical simulations. Wind represents a particularly challenging variable to model due to its high spatial and temporal variability. This paper presents a novel approach that integrates Gaussian processes and neural networks to model surface wind gusts at sub-kilometer resolution, leveraging multiple data sources, including numerical weather prediction models, topographical descriptors, and in-situ measurements. Results demonstrate the added value of modeling the multivariate covariance structure of the variable of interest, as opposed to only applying a univariate probabilistic regression approach. Modeling the covariance enables the optimal integration of observed measurements from ground stations, which is shown to reduce the continuous ranked probability score compared to the baseline. Moreover, it allows the generation of realistic fields that are also marginally calibrated, aided by scalable techniques such as random Fourier features and pathwise conditioning. We discuss the effect of different modeling choices, as well as different degrees of approximation, and present our results for a case study. © 2024, CC BY.

关键词： Wind forecasting

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：