Regularized kernel discriminant analysis (RKDA) performs linear discriminant analysis in the feature space via the kernel trick. Its performance depends on the selection of kernels. In this paper, we consider the problem of multiple kernel learning (MKL) for RKDA, in which the optimal kernel matrix is obtained as a linear combination of pre-specified kernel matrices. We show that the kernel learning problem in RKDA can be formulated as convex programs. First, we show that this problem can be formulated as a semidefinite program (SDP). Based on the equivalence relationship between RKDA and least square problems in the binary-class case, we propose a convex quadratically constrained quadratic programming (QCQP) formulation for kernel learning in RKDA. A semi-infinite linear programming (SILP) formulation is derived to further improve the efficiency. We extend these formulations to the multi-class case based on a key result established in this paper. That is, the multi-class RKDA kernel learning problem can be decomposed into a set of binary-class kernel learning problems which are constrained to share a common kernel. Based on this decomposition property, SDP formulations are proposed for the multi-class case. Furthermore, it leads naturally to QCQP and SILP formulations. As the performance of RKDA depends on the regularization parameter, we show that this parameter can also be optimized in a joint framework with the kernel. Extensive experiments have been conducted and analyzed, and connections to other algorithms are discussed.

We present a new approach to general-activity human pose estimation from depth images, building on Hough forests. We extend existing techniques in several ways: real time prediction of multiple 3D joints, explicit learning of voting weights, vote compression to allow larger training sets, and a comparison of several decision-tree training objectives. Key aspects of our work include: regression directly from the raw depth image, without the use of an arbitrary intermediate representation; applicability to general motions (not constrained to particular activities) and the ability to localize occluded as well as visible body joints. Experimental results demonstrate that our method produces state of the art results on several data sets including the challenging MSRC-5000 pose estimation test set, at a speed of about 200 frames per second. Results on silhouettes suggest broader applicability to other imaging modalities. 1.

Perhaps surprisingly, no practical performance models exist for popular (and complex) client applications such as Adobe’s Creative Suite, Microsoft’s Office and Visual Studio, Mozilla, Halo 3, etc. There is currently no tool that automatically answers program developers’, IT administrators ’ and end-users ’ simple what-if questions like “what happens to the performance of my favorite application X if I upgrade from Windows Vista to Windows 7?”. This paper describes our approach towards constructing practical, versatile performance models to address this problem. The goal is to have these models be useful for application developers to help expand application testing coverage and for IT administrators to assist with understanding the performance consequences of a software, hardware or configuration change. This paper’s main contributions are in system building and performance modeling. We believe we have built applications that are easier to model because we have proactively instrumented them to export their state and associated metrics. This application-specific monitoring is always on and interesting data is collected from real, "in-the-wild " deployments. The models we are experimenting with are based on statistical techniques. They require no modifications to the OS or applications beyond the above instrumentation, and no explicit a priori model on how an OS or application should behave. We are in the process of learning from models we have constructed for several Microsoft products, including the Office suite, Visual Studio and Media Player. This paper presents preliminary findings from a large user deployment (several hundred thousand user sessions) of these applications that show the coverage and limitations of such models. These findings pushed us to move beyond averages/means and go into some depth into why client application performance has an inherently large variance.

Active Appearance Models (AAMs) are widely used to fit shape models to new images. Recently it has been demonstrated that non-linear regression methods and sequences of AAMs can significantly improve performance over the original linear formulation. In this paper we focus on the ability of a model trained on one dataset to generalise to other sets with different conditions. In particular we compare two non-linear, discriminative regression strategies for predicting shape updates, a boosting approach and variants of Random Forest regression. We investigate the use of these regression methods within a sequential model fitting framework, where each stage in the sequence consists of a shape model and a corresponding regression model. The performance of the framework is assessed by both testing on unseen data taken from within the training databases, as well as by investigating the more difficult task of generalising to unrelated datasets. We present results that show that (a) the generalisation performance of the Random Forest is superior to that of the linear or boosted regression procedure and that (b) using a simple feature selection procedure, the Random Forest can be made to be as efficient as the boosting procedure without significant reduction in accuracy. 1

addressestd problem ofautC4WP ically selectKH tl best amongoutgvP from multW4v machine tevH+K+Hv3 (MT)systPC4 ExistPC approachesselect tl outct assigned ts highest score accordingt at arget language model. In some cases,ts existvH approaches donot work well. This paper proposest wo met4 ds t improve performance. Thefirst mett d is based on amult4HC comparisontom and checkswhetBL a score from language and tdvP4KWPv3 models is significantg highertgh t otgher The secondmetn d is based on probabilit ytHP at+KLBv3BLW isnot inferiort t otferi which ispredictv fromto above scores. Experimenti result showt hat t he proposed metH ds achieve an improvement f2t o 6 % in performance.

Isotonic regression is a well-known nonparametric tool for fitting monotonic models and has been studied from both theoretical and practical aspects for several decades, with applications in psychology, medicine, biology, among others. However, it has enjoyed only limited interest in recent years in the context of modern statistical applications. We believe the two major reasons for this limited attention are computational difficulties on large data and statistical difficulties (overfitting). We present a novel algorithmic approach to isotonic regression that addresses these concerns in a manner that is both practically useful and of independent methodological and algorithmic interest. Our new algorithm for isotonic regression is based on recursively partitioning the predictor space through solution of progressively smaller best cut subproblems. This creates a regularized sequence of isotonic models of increasing model complexity that converges to the global isotonic regression solution. The models along the sequence are often more accurate than the unregularized isotonic regression model because of the complexity control they offer. We offer quantification of this complexity control through estimation of degrees of freedom along the path. We also develop efficient methods for generating the global solution through this sequence of structured subproblems, as each subproblem is equivalent to a network flow problem for which efficient algorithms exist. The success of the regularized models in prediction and our algorithms favorable computational properties are demonstrated through a series of simulated and real data experiments.

Aggregated search is the task of integrating results from potentially multiple specialized search services, or verticals, into the Web search results. The task requires predicting not only which verticals to present (the focus of most prior research), but also predicting where in the Web results to present them (i.e., above or below the Web results, or somewhere in between). Learning models to aggregate results from multiple verticals is associated with two major challenges. First, because verticals retrieve different types of results and address different search tasks, results from different verticals are associated with different types of predictive evidence (or features). Second, even when a feature is common across verticals, its predictiveness may be verticalspecific. Therefore, approaches to aggregating vertical results require handling an inconsistent feature representation across verticals, and, potentially, a vertical-specific relationship between features and relevance. We present 3 general approaches that address these challenges in different ways and compare their results across a set of 13 verticals and 1070 queries. We show that the best approaches are those that allow the learning algorithm to learn a vertical-specific relationship between features and relevance.

discriminant analysis, principal component analysis Metagenomics is a primary tool for the description of microbial and viral communities. The sheer magnitude of the data generated in each metagenome makes identifying key differences in the function and taxonomy between communities difficult to elucidate. Here we discuss the application of seven different statistical analyses by comparing and contrasting the metabolic functions of 212 microbial metagenomes within and between 10 environments. Not all approaches are appropriate for all questions, and this work demonstrated the use of each approach. For example, Random Forests provided a robust and enlightening description of both the clustering of metagenomes and the metabolic processes that were important in separating microbial communities from different environments. All analysis identified that

Copyright © 2012 Allal Houssaïni et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Background. Many statistical models have been tested to predict phenotypic or virological response from genotypic data. A statistical framework called Super Learner has been introduced either to compare different methods/learners (discrete Super Learner) or to combine them in a Super Learner prediction method. Methods. The Jaguar trial is used to apply the Super Learner framework. The Jaguar study is an “add-on ” trial comparing the efficacy of adding didanosine to an on-going failing regimen. Our aim was also to investigate the impact on the use of different cross-validation strategies and different loss functions. Four different repartitions between training set and validations set were tested through two loss functions. Six statistical methods were compared. We assess performance by evaluating R 2 values and accuracy by calculating the rates of patients being correctly classified. Results. Our results indicated that the more recent Super Learner methodology of building a new predictor based on a weighted combination of different methods/learners provided good performance. A simple linear model provided similar results to those of this new predictor. Slight discrepancy arises between the two loss functions investigated, and slight difference arises also between