检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Kim, Gyuwan Li, Yang Spiliopoulou, Evangelia Ma, Jie Ballesteros, Miguel Wang, William Yang University of California Santa Barbara United States AWS AI Lab United States

The widespread deployment of large language models (LLMs) has led to impressive advancements, yet information about their training data, a critical factor in their performance, remains undisclosed. Membership inference attacks (MIAs) aim to determine whether a specific instance was part of a target model's training data. MIAs can offer insights into LLM outputs and help detect and address concerns such as data contamination and compliance with privacy and copyright standards. However, applying MIAs to LLMs presents unique challenges due to the massive scale of pre-training data and the ambiguous nature of membership. Additionally, creating appropriate benchmarks to evaluate MIA methods is not straightforward, as training and test data distributions are often unknown. In this paper, we introduce EM-MIA, a novel MIA method for LLMs that iteratively refines membership scores and prefix scores via an expectation-maximization algorithm, leveraging the duality that the estimates of these scores can be improved by each other. Membership scores and prefix scores assess how each instance is likely to be a member and discriminative as a prefix, respectively. Our method achieves state-of-the-art results on the WikiMIA dataset. To further evaluate EMMIA, we present OLMoMIA, a benchmark built from OLMo resources, which allows us to control the difficulty of MIA tasks with varying degrees of overlap between training and test data distributions. We believe that EM-MIA serves as a robust MIA method for LLMs and that OLMoMIA provides a valuable resource for comprehensively evaluating MIA approaches, thereby driving future research in this critical area. Copyright © 2024, The Authors. All rights reserved.

关键词： expectation maximization algorithm

来源：评论

学校读者我要写书评

暂无评论

Fast Identification of Transients: Applying expectation maximization to Neutrino Data

arXiv

引用

arXiv 2023年

作者： Karl, Martina Eller, Philipp Technische Universität München TUM School of Natural Sciences Physics Department James-Frank-Str. 1 Garching bei MünchenD-85748 Germany European Southern Observatory Karl-Schwarzschild-Str. 2 Garching bei MünchenD-85748 Germany

We present a novel method for identifying transients suitable for both strong signal-dominated and background-dominated objects. By employing the unsupervised machine learning algorithm known as expectation maximization, we achieve computing time reductions of over 104 on a single CPU compared to conventional brute-force methods. Furthermore, this approach can be readily extended to analyze multiple flares. We illustrate the algorithm’s application by fitting the IceCube neutrino flare of TXS 0506+056. Copyright © 2023, The Authors. All rights reserved.

关键词： expectation maximization algorithm

来源：评论

学校读者我要写书评

暂无评论

Theoretical and Numerical Study of MLEM and OSEM Reconstruction algorithms for Motion Correction in Emission Tomography

引用

IEEE TRANSACTIONS ON NUCLEAR SCIENCE 2009年第5期56卷 2739-2749页

作者： Dey, Joyoni King, Michael A. Univ Massachusetts Sch Med Div Nucl Med & Radiol Worcester MA 01655 USA

Patient body-motion and respiratory-motion impacts the image quality of cardiac SPECT and PET perfusion images. Several algorithms exist in the literature to correct for motion within the iterative maximum-likelihood reconstruction framework. In this work. three algorithms are derived starting with Poisson statistics to correct for patient motion. The first one is a motion compensated MLEM algorithm (MC-MLEM). The next two algorithms called MGEM-1 and MGEM-2 (short for Motion Gated OSEM, 1 and 2) use the motion states as subsets, in two different ways. Experiments were performed with NCAT phantoms (with exactly known motion) as the source and attenuation distributions. Experiments were also performed on in anthropomorphic phantom and a patient study. The SIMIND Monte Carlo simulation software was used to create SPECT projection images of the NCAT phantoms. The projection images were then modified to have Poisson noise levels equivalent to that of clinical acquisition. We investigated application of these algorithms to correction of (1) 1 large body-motion of 2 cm in Superior-Inferior (SI) and Anterior-Posterior (AP) directions each and (2) respiratory motion of 2 cm in SI and 0.6 cm in AP. We determined the bias with respect to the NCAT phantom activity for noiseless reconstructions as well as the bias-variance for noisy reconstructions. The MGEM-1 advanced along the bias-variance curve faster than the MC-MLEIM with iterations. The MCEM-1 also lowered the noiseless bias (with respect to NCAT truth) faster with iterations, compared to file MC-MLEM algorithms, as expected with subset algorithms. For the body motion correction with two motion states, after the 9th iteration the Was was close to that of MC-MLEM at iteration 17, reducing the number of iterations by a factor of 1.89. For the respiratory motion correction with 9 motion states, based on the noiseless bias, the iteration reduction factor was approximately 7. For the MGEM-2, however. bias-plot or the bias-var

关键词： expectation maximization algorithm image reconstruction

来源：评论

学校读者我要写书评

暂无评论

Bayesian methods for predicting interacting protein pairs using domain information

引用

BIOMETRICS 2007年第3期63卷 824-833页

作者： Kim, Inyoung Liu, Yin Zhao, Hongyu Yale Univ Dept Epidemiol & Publ Hlth Sch Med New Haven CT 06520 USA Yale Univ Program Computat Biol & Bioinformat New Haven CT 06520 USA Yale Univ Sch Med Dept Genet New Haven CT 06520 USA

Protein-protein interactions (PPIs) play important roles in most fundamental cellular processes including cell cycle, metabolism, and cell proliferation. Therefore, the development of effective statistical approaches to predicting protein interactions based on recently available large-scale experimental data is very important. Because protein domains are the functional units of proteins and PPIs are mostly achieved through domain-domain interactions (DDIs), the modeling and analysis of protein interactions at the domain level may be more informative and insightful. However, due to the large number of domains, the number of parameters to be estimated is very large, yet the amount of information for statistical inference is quite limited. In this article we propose a full Bayesian method and a semi-Bayesian method for simultaneously estimating DDI probabilities, the false positive rate, and the false negative rate of high-throughput data through integrating data from several organisms. We also propose a model to associate protein interaction probabilities with domain interaction probabilities that reflects the number of domains in each protein, Our Bayesian methods are compared with the likelihood-based approach (Deng et al., 2002, Genome Research 12, 1504-1508;Liu, Liu, and Zhao, 2005, Bioinformatics 21, 3279-3285) developed using the expectation maximization algorithm. We show that the full Bayesian method has the smallest mean square error through both simulations and theoretical justification under a special scenario. The large-scale PPI data obtained from high-throughput yeast two-hybrid experiments are used to demonstrate the advantages of the Bayesian approaches.

关键词： Bayesian method domain-domain interaction expectation maximization algorithm protein-protein interaction

来源：评论

学校读者我要写书评

暂无评论

Anomaly detection of structural health monitoring data using the maximum likelihood estimation-based Bayesian dynamic linear model

引用

STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL 2021年第6期20卷 2936-2952页

作者： Zhang, Yi-Ming Wang, Hao Wan, Hua-Ping Mao, Jian-Xiao Xu, Yi-Chao Southeast Univ Minist Educ Key Lab C&PC Struct Nanjing 211189 Peoples R China Monash Univ Dept Civil Engn Clayton Vic Australia Zhejiang Univ Coll Civil Engn & Architecture Hangzhou Peoples R China Jiangsu Transportat Res Inst Co Ltd Nanjing Peoples R China State Key Lab Safety & Hlth In Serv Long Span Bri Nanjing Peoples R China

Enormous data are continuously collected by the structural health monitoring system of civil infrastructures. The structural health monitoring data inevitably involve anomalies caused by sensors, transmission errors, or abnormal structural behaviors. It is important to identify the anomalies and find their origin (e.g. sensor fault or structural damage) to make correct interventions. Moreover, online anomaly identification of the structural health monitoring data is critical for timely structural condition assessment and decision-making. This study proposes an online approach for detecting anomalies of the structural health monitoring data based on the Bayesian dynamic linear model. In particular, Bayesian dynamic linear model, consisting of various components, is implemented to characterize the feature of real-time measurements. expectation maximization algorithm and Kalman smoother are combined to estimate the Bayesian dynamic linear model parameters and generate log-likelihood functions. The subspace identification method is introduced to overcome the initialization issue of the expectation maximization algorithm. The log-likelihood difference of consecutive time steps is then used to determine thresholds without introducing extra anomaly detectors. The proposed Bayesian dynamic linear model-based approach is first illustrated by the simulation data and then applied to the structural health monitoring data collected from two long-span bridges. The results indicate that the proposed method exhibits good accuracy and high computational efficiency and also allows for reconstructing the strain measurements to replace anomalies.

关键词： Bayesian dynamic linear model anomaly identification structural health monitoring log-likelihood expectation maximization algorithm

来源：评论

学校读者我要写书评

暂无评论

Estimating the modal parameters from multiple measurement setups using a joint state space model

引用

MECHANICAL SYSTEMS AND SIGNAL PROCESSING 2014年第1-2期43卷 171-191页

作者： Javier Cara, F. Juan, Jesus Alarcon, Enrique Univ Politecn Madrid ETS Ingenieros Ind Stat Lab E-28006 Madrid Spain Univ Politecn Madrid ETS Ingenieros Ind Dept Struct Mech E-28006 Madrid Spain

Computing the modal parameters of structural systems often requires processing data from multiple non-simultaneously recorded setups of sensors. These setups share some sensors in common, the so-called reference sensors, which are fixed for all measurements, while the other sensors change their position from one setup to the next. One possibility is to process the setups separately resulting in different modal parameter estimates for each setup. Then, the reference sensors are used to merge or glue the different parts of the mode shapes to obtain global mode shapes, while the natural frequencies and damping ratios are usually averaged. In this paper we present a new state space model that processes all setups at once. The result is that the global mode shapes are obtained automatically, and only a value for the natural frequency and damping ratio of each mode is estimated. We also investigate the estimation of this model using maximum likelihood and the expectation maximization algorithm, and apply this technique to simulated and measured data corresponding to different structures. (C) 2013 Elsevier Ltd. All rights reserved.

关键词： Assembled mode shapes Multi-setup operational modal analysis Maximum likelihood estimation expectation maximization algorithm State space model

来源：评论

学校读者我要写书评

暂无评论

Robust algorithm for attack detection based on time-varying hidden Markov model subject to outliers

引用

INTERNATIONAL JOURNAL OF ADAPTIVE CONTROL AND SIGNAL PROCESSING 2020年第10期34卷 1537-1558页

作者： Lu, Genghong Feng, Dongqin Huang, Biao Zhejiang Univ Inst Cyber Syst & Control State Key Lab Ind Control Technol Zheda Rd 38 Hangzhou Zhejiang Peoples R China Univ Alberta Dept Chem & Mat Engn Edmonton AB Canada

The problem of robust attack detection and prediction for networked control systems in the presence of outliers is discussed in this article. The conventional hidden Markov model (HMM) is trained to learn the system behavior (ie, transitions between different operating modes) in the nominal process. The HMM with time-varying transition probabilities is used to track the attack behavior in which the adversary triggers more hazard modes to hasten fatigue of control devices by injecting attack signals with random magnitude and frequency. For different operating modes, the observations are assumed to follow different multivariate Student'stdistributions instead of Gaussian distributions and thus address the robust estimation problem. The expectation maximization algorithm is used to estimate parameters. Finally, simulations are conducted to verify the effectiveness of the proposed method.

关键词： expectation maximization algorithm hidden Markov models outliers robust attack detection time-varying transition probabilities

来源：评论

学校读者我要写书评

暂无评论

Probabilistic modeling and recognition of 3-D objects

引用

INTERNATIONAL JOURNAL OF COMPUTER VISION 2000年第3期39卷 229-251页

作者： Hornegger, J Niemann, H Univ Erlangen Nurnberg Lehrstuhl Mustererkennung Informat 5 D-91058 Erlangen Germany

This paper introduces a uniform statistical framework for both 3-D and 2-D object recognition using intensity images as input data. The theoretical part provides a mathematical tool for stochastic modeling. The algorithmic part introduces methods for automatic model generation, localization, and recognition of objects. 2-D images are used for learning the statistical appearance of 3-D objects;both the depth information and the matching between image and model features are missing for model generation. The implied incomplete data estimation problem is solved by the expectation maximization algorithm. This leads to a novel class of algorithms for automatic model generation from projections. The estimation of pose parameters corresponds to a non-linear maximum likelihood estimation problem which is solved by a global optimization procedure. Classification is done by the Bayesian decision rule. This work includes the experimental evaluation of the various facets of the presented approach. An empirical evaluation of learning algorithms and the comparison of different pose estimation algorithms show the feasibility of the proposed probabilistic framework.

关键词： statistical object recognition pose estimation expectation maximization algorithm mixture densities hidden Markov models marginalization global optimization adaptive random search

来源：评论

学校读者我要写书评

暂无评论

Film hole layout optimization for multi-row film cooling plate in continuous parameter space under non-uniform inlet temperature distribution

引用

APPLIED THERMAL ENGINEERING 2024年 257卷

作者： Huang, Junjie Zhu, Jianqin Cheng, Zeyuan Tong, Zixiang Qiu, Lu Wang, Yanjia Beihang Univ Sch Energy & Power Engn Natl Key Lab Sci & Technol Aeroengine Aerothermody Beijing 100191 Peoples R China Tianmushan Lab Hangzhou 310023 Peoples R China

Film cooling is an advanced external cooling technology for protecting gas turbine blades from excessive temperatures. To counteract non-uniform heat loads caused by hot streak and swirl effects at the combustion chamber outlet, fine design of film hole positions is necessary. However, the complexity due to the numerous film holes and their positional parameters presents challenges in the independent optimization of film hole positions in a continuous parameter space. This study proposed an optimization method grounded in the divide and conquer approach to address optimization for film hole layout. The film hole layout was decomposed into multiple single rows and optimized systematically by iteratively refining the single-row film hole layout along the streamwise direction using the expectation maximization algorithm. Optimized layouts for five different inlet temperature distributions were obtained using the proposed method and adiabatic wall temperature distributions were compared and analyzed for optimized and staggered layouts. Additionally, the cooling performance of the optimized layout was evaluated against the staggered layout under conjugate heat transfer conditions and the robustness of the optimized layout under various blowing ratios and peak temperature positions was tested. The results demonstrated that the proposed method can optimize the positions of 52 film holes within 0.3 h, and it was generalized to various inlet temperature distributions. By enhancing lateral interactions among downstream film jets, the lifting effect of kidney vortex in the optimized layout was weakened, bringing the film jets closer to the wall and significantly increasing coolant coverage during streamwise development. Comparative analysis reveals the superior cooling performance of the optimized layout at the blowing ratio of 1.0, evidenced by a 15.1% reduction in total input heat transfer rate and a 12.3% enhancement in overall cooling efficiency relative to the staggered l

关键词： Non-uniform temperature distribution Film cooling Hole layout optimization Divide and conquer approach expectation maximization algorithm

来源：评论

学校读者我要写书评

暂无评论

Classical and Bayesian Estimation for the Parameters of a Competing Risk Model Based on Minimum of Exponential and Gamma Failures

引用

IEEE TRANSACTIONS ON RELIABILITY 2016年第3期65卷 1522-1535页

作者： Ranjan, Rakesh Upadhyay, Satyanshu K. Banaras Hindu Univ Dept Stat Varanasi 221005 Uttar Pradesh India Banaras Hindu Univ DST Ctr Interdisciplinary Math Sci Varanasi 221005 Uttar Pradesh India

The paper provides both classical and Bayesian estimation of the parameters of a competing risk model defined on the basis of minimum of exponential and gamma failure modes. Usually such situations are the examples of incomplete specification of data that naturally opens the way to expectation maximization algorithm for obtaining maximum likelihood estimates of model parameters. This incomplete specification of the data simultaneously explores the possibility of sampling importance resampling strategy with intermediate Markov chain Monte Carlo steps for the Bayesian estimation of parameters. Although this paper focuses primarily on estimation of model parameters, other inferential developments can be routinely done. Numerical illustration is provided based on both simulated and real-data examples.

关键词： Competing risk model exponential model expectation maximization algorithm gamma model increasing hazard rate sampling importance resampling

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

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

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：