检索结果-内蒙古大学图书馆

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Jun Zhu Ning Chen Eric P. Xing Google Department of Computer Science and Technology State Key Laboratory of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science and Technology Tsinghua University Beijing China School of Computer Science Carnegie Mellon University Pittsburgh PA

Existing Bayesian models, especially nonparametric Bayesian methods, rely on specially conceived priors to incorporate domain knowledge for discovering improved latent representations. While priors affect posterior distributions through Bayes' rule, imposing posterior regularization is arguably more direct and in some cases more natural and general. In this paper, we present regularized Bayesian inference (RegBayes), a novel computational framework that performs posterior inference with a regularization term on the desired post-data posterior distribution under an information theoretical formulation. RegBayes is more flexible than the procedure that elicits expert knowledge via priors, and it covers both directed Bayesian networks and undirected Markov networks. When the regularization is induced from a linear operator on the posterior distributions, such as the expectation operator, we present a general convex-analysis theorem to characterize the solution of RegBayes. Furthermore, we present two concrete examples of RegBayes, infinite latent support vector machines (iLSVM) and multi-task infinite latent support vector machines (MT-iLSVM), which explore the large-margin idea in combination with a nonparametric Bayesian model for discovering predictive latent features for classification and multi-task learning, respectively. We present efficient inference methods and report empirical studies on several benchmark data sets, which appear to demonstrate the merits inherited from both large-margin learning and Bayesian nonparametrics. Such results contribute to push forward the interface between these two important subfields, which have been largely treated as isolated in the community.

关键词： Bayesian inference Bayesian nonparametrics classification large-margin learning multi-task learning posterior regularization

来源：评论

学校读者我要写书评

暂无评论

A scalable approach to column-based low-rank matrix approximation

A scalable approach to column-based low-rank matrix approxim...

引用

23rd International Joint Conference on Artificial Intelligence, IJCAI 2013

作者： Pi, Yifan Peng, Haoruo Zhou, Shuchang Zhang, Zhihua Institute for Theoretical Computer Science IIIS Tsinghua University Beijing 100084 China Department of Computer Science and Technology Tsinghua University Beijing 100084 China State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences Beijing 100190 China Google Inc. Beijing 100084 China College of Computer Science and Technology Zhejiang University Hangzhou 310027 China

ISBN: (纸本)9781577356332

In this paper, we address the column-based low-rank matrix approximation problem using a novel parallel approach. Our approach is based on the divide-and-combine idea. We first perform column selection on submatrices of an original data matrix in parallel, and then combine the selected columns into the final output. Our approach enjoys a theoretical relative-error upper bound. In addition, our column-based low-rank approximation partitions data in a deterministic way and makes no assumptions about matrix coherence. Compared with other traditional methods, our approach is scalable on largescale matrices. Finally, experiments on both simulated and real world data show that our approach is both efficient and effective.

关键词： Matrix algebra

来源：评论

学校读者我要写书评

暂无评论

How long will she call me? Distribution, social theory and duration prediction

How long will she call me? Distribution, social theory and d...

引用

European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2013

作者： Dong, Yuxiao Tang, Jie Lou, Tiancheng Wu, Bin Chawla, Nitesh V. Department of Computer Science and Engineering University of Notre Dame United States Department of Computer Science and Technology Tsinghua University China Google Inc. United States Beijing University of Posts and Telecommunications China Interdisciplinary Center for Network Science and Applications University of Notre Dame United States

ISBN: (纸本)9783642409905

Call duration analysis is a key issue for understanding underlying patterns of (mobile) phone users. In this paper, we study to which extent the duration of a call between users can be predicted in a dynamic mobile network. We have collected a mobile phone call data from a mobile operating company, which results in a network of 272,345 users and 3.9 million call records during two months. We first examine the dynamic distribution properties of the mobile network including periodicity and demographics. Then we study several important social theories in the call network including strong/weak ties, link homophily, opinion leader and social balance. The study reveals several interesting phenomena such as people with strong ties tend to make shorter calls and young females tend to make long calls, in particular in the evening. Finally, we present a time-dependent factor graph model to model and infer the call duration between users, by incorporating our observations in the distribution analysis and the social theory analysis. Experiments show that the presented model can achieve much better predictive performance compared to several baseline methods. Our study offers evidences for social theories and also unveils several different patterns in the call network from online social networks. © 2013 Springer-Verlag.

关键词： Wireless networks

来源：评论

学校读者我要写书评

暂无评论

Active learning using smooth relative regret approximations with applications

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Nir Ailon Ron Begleiter Esther Ezra Google Department of Computer Science Technion Israel Institute of Technology Haifa Israel Coutrant Institute of Mathematical Science New York University New York NY

The disagreement coeffcient of Hanneke has become a central data independent invariant in proving active learning rates. It has been shown in various ways that a concept class with low complexity together with a bound on the disagreement coeffcient at an optimal solution allows active learning rates that are superior to passive learning *** present a different tool for pool based active learning which follows from the existence of a certain uniform version of low disagreement coeffcient, but is not equivalent to it. In fact, we present two fundamental active learning problems of significant interest for which our approach allows nontrivial active learning bounds. However, any general purpose method relying on the disagreement coeffcient bounds only, fails to guarantee any useful bounds for these problems. The applications of interest are: Learning to rank from pairwise preferences, and clustering with side information (a.k.a. semi-supervised clustering).The tool we use is based on the learner's ability to compute an estimator of the difference between the loss of any hypothesis and some fixed "pivotal" hypothesis to within an absolute error of at most ε times the disagreement measure (l1 distance) between the two hypotheses. We prove that such an estimator implies the existence of a learning algorithm which, at each iteration, reduces its in-class excess risk to within a constant factor. Each iteration replaces the current pivotal hypothesis with the minimizer of the estimated loss difference function with respect to the previous pivotal hypothesis. The label complexity essentially becomes that of computing this estimator.

关键词： active learning clustering with side information disagreement coeffcient learning to rank from pairwise preferences semi-supervised clustering smooth relative regret approximation

来源：评论

学校读者我要写书评

暂无评论

Efficient state-space inference of periodic latent force models

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Steven Reece Siddhartha Ghosh Alex Rogers Stephen Roberts Nicholas R. Jennings Google Department of Engineering Science University of Oxford Oxford UK Electronics and Computer Science University of Southampton Southampton UK Electronics and Computer Science University of Southampton Southampton UK and Department of Computing and Information Technology King Abdulaziz University Saudi Arabia

Latent force models (LFM) are principled approaches to incorporating solutions to differential equations within non-parametric inference methods. Unfortunately, the development and application of LFMs can be inhibited by their computational cost, especially when closed-form solutions for the LFM are unavailable, as is the case in many real world problems where these latent forces exhibit periodic behaviour. Given this, we develop a new sparse representation of LFMs which considerably improves their computational efficiency, as well as broadening their applicability, in a principled way, to domains with periodic or near periodic latent forces. Our approach uses a linear basis model to approximate one generative model for each periodic force. We assume that the latent forces are generated from Gaussian process priors and develop a linear basis model which fully expresses these priors. We apply our approach to model the thermal dynamics of domestic buildings and show that it is effective at predicting day-ahead temperatures within the homes. We also apply our approach within queueing theory in which quasi-periodic arrival rates are modelled as latent forces. In both cases, we demonstrate that our approach can be implemented efficiently using state-space methods which encode the linear dynamic systems via LFMs. Further, we show that state estimates obtained using periodic latent force models can reduce the root mean squared error to 17% of that from non-periodic models and 27% of the nearest rival approach which is the resonator model (Särkkä et al., 2012; Hartikainen et al., 2012).

关键词： Gaussian processes Kalman filter kernel principle component analysis latent force models queueing theory

来源：评论

学校读者我要写书评

暂无评论

Reinforcement learning for closed-loop propofol anesthesia: a study in human volunteers

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Brett L. Moore Larry D. Pyeatt Vivekanand Kulkarni Periklis Panousis Kevin Padrez Anthony G. Doufas Google Department of Computer Science Texas Tech University Lubbock TX Department of Mathematics and Computer Science South Dakota School of Mines and Technology Rapid City SD Department of Anesthesiology Perioperative and Pain Medicine Stanford University School of Medicine Stanford CA

Clinical research has demonstrated the efficacy of closed-loop control of anesthesia using the bispectral index of the electroencephalogram as the controlled variable. These controllers have evolved to yield patient-specific anesthesia, which is associated with improved patient outcomes. Despite progress, the problem of patient-specific anesthesia remains unsolved. A variety of factors confound good control, including variations in human physiology, imperfect measures of drug effect, and delayed, hysteretic response to drug delivery. Reinforcement learning (RL) appears to be uniquely equipped to overcome these challenges; however, the literature offers no precedent for RL in anesthesia. To begin exploring the role RL might play in improving anesthetic care, we investigated the method's application in the delivery of patient-specific, propofol-induced hypnosis in human volunteers. When compared to performance metrics reported in the anesthesia literature, RL demonstrated patient-specific control marked by improved accuracy and stability. Furthermore, these results suggest that RL may be considered a viable alternative for solving other difficult closed-loop control problems in medicine. More rigorous clinical study, beyond the confines of controlled human volunteer studies, is needed to substantiate these findings.

关键词： anesthesia bispectral index closed-loop control hypnosis propofol reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Parallelizing exploration-exploitation tradeoffs in Gaussian process bandit optimization

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Thomas Desautels Andreas Krause Joel W. Burdick Google University College London London UK Department of Computer Science ETH Zurich Zürich Switzerland Department of Mechanical Engineering California Institute of Technology Pasadena CA

How can we take advantage of opportunities for experimental parallelization in exploration-exploitation tradeoffs? In many experimental scenarios, it is often desirable to execute experiments simultaneously or in batches, rather than only performing one at a time. Additionally, observations may be both noisy and expensive. We introduce Gaussian Process Batch Upper Confidence Bound (GP-BUCB), an upper confidence bound-based algorithm, which models the reward function as a sample from a Gaussian process and which can select batches of experiments to run in parallel. We prove a general regret bound for GP-BUCB, as well as the surprising result that for some common kernels, the asymptotic average regret can be made independent of the batch size. The GP-BUCB algorithm is also applicable in the related case of a delay between initiation of an experiment and observation of its results, for which the same regret bounds hold. We also introduce Gaussian Process Adaptive Upper Confidence Bound (GP-AUCB), a variant of GP-BUCB which can exploit parallelism in an adaptive manner. We evaluate GP-BUCB and GP-AUCB on several simulated and real data sets. These experiments show that GP-BUCB and GP-AUCB are competitive with state-of-the-art heuristics.

关键词： Gaussian process active learning batch regret bound upper confidence bound

来源：评论

学校读者我要写书评

暂无评论

On the bayes-optimality of F-measure maximizers

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Willem Waegeman Krzysztof Dembczyńki Arkadiusz Jachnik Weiwei Cheng Eyke Hüllermeier Google Department of Mathematical Modelling Statistics and Bioinformatics Ghent University Ghent Belgium Institute of Computing Science Poznan University of Technology Poznan Poland Amazon Development Center Germany Berlin Germany Department of Computer Science University of Paderborn Paderborn Germany

The F-measure, which has originally been introduced in information retrieval, is nowadays routinely used as a performance metric for problems such as binary classification, multi-label classification, and structured output prediction. Optimizing this measure is a statistically and computationally challenging problem, since no closed-form solution exists. Adopting a decision-theoretic perspective, this article provides a formal and experimental analysis of different approaches for maximizing the F-measure. We start with a Bayes-risk analysis of related loss functions, such as Hamming loss and subset zero-one loss, showing that optimizing such losses as a surrogate of the F-measure leads to a high worst-case regret. Subsequently, we perform a similar type of analysis for F-measure maximizing algorithms, showing that such algorithms are approximate, while relying on additional assumptions regarding the statistical distribution of the binary response variables. Furthermore, we present a new algorithm which is not only computationally efficient but also Bayes-optimal, regardless of the underlying distribution. To this end, the algorithm requires only a quadratic (with respect to the number of binary responses) number of parameters of the joint distribution. We illustrate the practical performance of all analyzed methods by means of experiments with multi-label classification problems.

关键词： F-measure bayes-optimal predictions multi-label classification regret statistical decision theory structured output prediction

来源：评论

学校读者我要写书评

暂无评论

Particle gibbs with ancestor sampling

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Fredrik Lindsten Michael I. Jordan Thomas B. Schön Google Department of Engineering University of Cambridge Cambridge UK and Division of Automatic Control Linköping University Linköping Sweden Computer Science Division and Department of Statistics University of California Berkeley CA Department of Information Technology Uppsala University Uppsala Sweden

Particle Markov chain Monte Carlo (PMCMC) is a systematic way of combining the two main tools used for Monte Carlo statistical inference: sequential Monte Carlo (SMC) and Markov chain Monte Carlo (MCMC). We present a new PMCMC algorithm that we refer to as particle Gibbs with ancestor sampling (PGAS). PGAS provides the data analyst with an off-the-shelf class of Markov kernels that can be used to simulate, for instance, the typically high-dimensional and highly autocorrelated state trajectory in a state-space model. The ancestor sampling procedure enables fast mixing of the PGAS kernel even when using seemingly few particles in the underlying SMC sampler. This is important as it can significantly reduce the computational burden that is typically associated with using SMC. PGAS is conceptually similar to the existing PG with backward simulation (PGBS) procedure. Instead of using separate forward and backward sweeps as in PGBS, however, we achieve the same effect in a single forward sweep. This makes PGAS well suited for addressing inference problems not only in state-space models, but also in models with more complex dependencies, such as non-Markovian, Bayesian nonparametric, and general probabilistic graphical models.

关键词： Bayesian inference non-Markovian models particle Markov chain Monte Carlo sequential Monte Carlo state-space models

来源：评论

学校读者我要写书评

暂无评论

Detecting click fraud in online advertising: a data mining approach

The Journal of Machine Learning Research

引用

The Journal of Machine Learning Research 2014年第1期15卷

作者： Kevin Murphy Bernhard Schölkopf Richard Oentaryo Ee-Peng Lim Michael Finegold David Lo Feida Zhu Clifton Phua Eng-Yeow Cheu Ghim-Eng Yap Kelvin Sim Minh Nhut Nguyen Kasun Perera Bijay Neupane Mustafa Faisal Zeyar Aung Wei Lee Woon Wei Chen Dhaval Patel Daniel Berrar Google Living Analytics Research Centre Singapore Management University Singapore SAS Institute Pte. Ltd. Singapore Data Analytics Department Institute for Infocomm Research Singapore Masdar Institute of Science and Technology Abu Dhabi United Arab Emirates Institute for Infocomm Research Singapore Department of Computer Science and Engineering Indian Institute of Technology Roorkee Roorkee Uttarakhand India Interdisciplinary Graduate School of Science and Engineering Tokyo Institute of Technology Yokohama Japan

Click fraud-the deliberate clicking on advertisements with no real interest on the product or service offered-is one of the most daunting problems in online advertising. Building an effective fraud detection method is thus pivotal for online advertising businesses. We organized a Fraud Detection in Mobile Advertising (FDMA) 2012 Competition, opening the opportunity for participants to work on real-world fraud data from BuzzCity Pte. Ltd., a global mobile advertising company based in Singapore. In particular, the task is to identify fraudulent publishers who generate illegitimate clicks, and distinguish them from normal publishers. The competition was held from September 1 to September 30, 2012, attracting 127 teams from more than 15 countries. The mobile advertising data are unique and complex, involving heterogeneous information, noisy patterns with missing values, and highly imbalanced class distribution. The competition results provide a comprehensive study on the usability of data mining-based fraud detection approaches in practical setting. Our principal findings are that features derived from fine-grained time-series analysis are crucial for accurate fraud detection, and that ensemble methods offer promising solutions to highly-imbalanced nonlinear classification tasks with mixed variable types and noisy/missing patterns. The competition data remain available for further studies at http://***/fdma2012/.

关键词： ensemble learning feature engineering fraud detection imbalanced classification

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：