# Showing papers in "Journal of Machine Learning Research in 2001"

••

Microsoft

^{1}TL;DR: It is demonstrated that by exploiting a probabilistic Bayesian learning framework, the 'relevance vector machine' (RVM) can derive accurate prediction models which typically utilise dramatically fewer basis functions than a comparable SVM while offering a number of additional advantages.

Abstract: This paper introduces a general Bayesian framework for obtaining sparse solutions to regression and classification tasks utilising models linear in the parameters Although this framework is fully general, we illustrate our approach with a particular specialisation that we denote the 'relevance vector machine' (RVM), a model of identical functional form to the popular and state-of-the-art 'support vector machine' (SVM) We demonstrate that by exploiting a probabilistic Bayesian learning framework, we can derive accurate prediction models which typically utilise dramatically fewer basis functions than a comparable SVM while offering a number of additional advantages These include the benefits of probabilistic predictions, automatic estimation of 'nuisance' parameters, and the facility to utilise arbitrary basis functions (eg non-'Mercer' kernels) We detail the Bayesian framework and associated learning algorithm for the RVM, and give some illustrative examples of its application along with some comparative benchmarks We offer some explanation for the exceptional degree of sparsity obtained, and discuss and demonstrate some of the advantageous features, and potential extensions, of Bayesian relevance learning

5,116 citations

••

TL;DR: A general method for combining the classifiers generated on the binary problems is proposed, and a general empirical multiclass loss bound is proved given the empirical loss of the individual binary learning algorithms.

Abstract: We present a unifying framework for studying the solution of multiclass categorization problems by reducing them to multiple binary problems that are then solved using a margin-based binary learning algorithm. The proposed framework unifies some of the most popular approaches in which each class is compared against all others, or in which all pairs of classes are compared to each other, or in which output codes with error-correcting properties are used. We propose a general method for combining the classifiers generated on the binary problems, and we prove a general empirical multiclass loss bound given the empirical loss of the individual binary learning algorithms. The scheme and the corresponding bounds apply to many popular classification learning algorithms including support-vector machines, AdaBoost, regression, logistic regression and decision-tree algorithms. We also give a multiclass generalization error analysis for general output codes with AdaBoost as the binary learner. Experimental results with SVM and AdaBoost show that our scheme provides a viable alternative to the most commonly used multiclass algorithms.

1,949 citations

••

TL;DR: Keywords: learning Reference EPFL-REPORT-82604 URL: http://publications.idiap.ch/downloads/reports/2000/rr00-17.pdf

Abstract: Support Vector Machines (SVMs) for regression problems are trained by solving a quadratic optimization problem which needs on the order of l square memory and time resources to solve, where l is the number of training examples. In this paper, we propose a decomposition algorithm, SVMTorch (available at http://www.idiap.ch/learning/SVMTorch.html ), which is similar to SVM-Light proposed by Joachims (1999) for classification problems, but adapted to regression problems. With this algorithm, one can now efficiently solve large-scale regression problems (more than 20000 examples). Comparisons with Nodelib, another publicly available SVM algorithm for large-scale regression problems from Flake and Lawrence (2000) yielded significant time improvements. Finally, based on a recent paper from Lin (2000), we show that a convergence proof exists for our algorithm.

829 citations

••

TL;DR: The mixtures-of-trees model as mentioned in this paper generalizes the probabilistic trees of Chow and Liu (1968) in a different and complementary direction to that of Bayesian networks.

Abstract: This paper describes the mixtures-of-trees model, a probabilistic model for discrete multidimensional domains. Mixtures-of-trees generalize the probabilistic trees of Chow and Liu (1968) in a different and complementary direction to that of Bayesian networks. We present efficient algorithms for learning mixtures-of-trees models in maximum likelihood and Bayesian frameworks. We also discuss additional efficiencies that can be obtained when data are "sparse," and we present data structures and algorithms that exploit such sparseness. Experimental results demonstrate the performance of the model for both density estimation and classification. We also discuss the sense in which tree-based classifiers perform an implicit form of feature selection, and demonstrate a resulting insensitivity to irrelevant attributes.

801 citations

••

TL;DR: An implicit Lagrangian for the dual of a simple reformulation of the standard quadratic program of a linear support vector machine is proposed, which leads to the minimization of an unconstrained differentiable convex function in a space of dimensionality equal to the number of classified points.

Abstract: An implicit Lagrangian for the dual of a simple reformulation of the standard quadratic program of a linear support vector machine is proposed. This leads to the minimization of an unconstrained differentiable convex function in a space of dimensionality equal to the number of classified points. This problem is solvable by an extremely simple linearly convergent Lagrangian support vector machine (LSVM) algorithm. LSVM requires the inversion at the outset of a single matrix of the order of the much smaller dimensionality of the original input space plus one. The full algorithm is given in this paper in 11 lines of MATLAB code without any special optimization tools such as linear or quadratic programming solvers. This LSVM code can be used "as is" to solve classification problems with millions of points. For example, 2 million points in 10 dimensional input space were classified by a linear surface in 82 minutes on a Pentium III 500 MHz notebook with 384 megabytes of memory (and additional swap space), and in 7 minutes on a 250 MHz UltraSPARC II processor with 2 gigabytes of memory. Other standard classification test problems were also solved. Nonlinear kernel classification can also be solved by LSVM. Although it does not scale up to very large problems, it can handle any positive semidefinite kernel and is guaranteed to converge. A short MATLAB code is also given for nonlinear kernels and tested on a number of problems.

634 citations

••

Microsoft

^{1}TL;DR: This work describes a graphical model for probabilistic relationships--an alternative to the Bayesian network--called a dependency network and identifies several basic properties of this representation and describes a computationally efficient procedure for learning the graph and probability components from data.

Abstract: We describe a graphical model for probabilistic relationships--an alternative to the Bayesian network--called a dependency network. The graph of a dependency network, unlike a Bayesian network, is potentially cyclic. The probability component of a dependency network, like a Bayesian network, is a set of conditional distributions, one for each node given its parents. We identify several basic properties of this representation and describe a computationally efficient procedure for learning the graph and probability components from data. We describe the application of this representation to probabilistic inference, collaborative filtering (the task of predicting preferences), and the visualization of acausal predictive relationships.

602 citations

••

[...]

TL;DR: It is found that Bayes point machines consistently outperform support vector machines on both surrogate data and real-world benchmark data sets and it is demonstrated that the real-valued output of single Bayes points on novel test points is a valid confidence measure and leads to a steady decrease in generalisation error when used as a rejection criterion.

Abstract: Kernel-classifiers comprise a powerful class of non-linear decision functions for binary classification. The support vector machine is an example of a learning algorithm for kernel classifiers that singles out the consistent classifier with the largest margin, i.e. minimal real-valued output on the training sample, within the set of consistent hypotheses, the so-called version space. We suggest the Bayes point machine as a well-founded improvement which approximates the Bayes-optimal decision by the centre of mass of version space. We present two algorithms to stochastically approximate the centre of mass of version space: a billiard sampling algorithm and a sampling algorithm based on the well known perceptron algorithm. It is shown how both algorithms can be extended to allow for soft-boundaries in order to admit training errors. Experimentally, we find that - for the zero training error case - Bayes point machines consistently outperform support vector machines on both surrogate data and real-world benchmark data sets. In the soft-boundary/soft-margin case, the improvement over support vector machines is shown to be reduced. Finally, we demonstrate that the real-valued output of single Bayes points on novel test points is a valid confidence measure and leads to a steady decrease in generalisation error when used as a rejection criterion.

243 citations

••

TL;DR: This paper develops the methodology for lifting known static bounds to the shifting case and obtains bounds when the comparison class consists of linear neurons (linear combinations of experts).

Abstract: In most on-line learning research the total on-line loss of the algorithm is compared to the total loss of the best off-line predictor u from a comparison class of predictors. We call such bounds static bounds. The interesting feature of these bounds is that they hold for an arbitrary sequence of examples. Recently some work has been done where the predictor ut at each trial t is allowed to change with time, and the total on-line loss of the algorithm is compared to the sum of the losses of ut at each trial plus the total "cost" for shifting to successive predictors. This is to model situations in which the examples change over time, and different predictors from the comparison class are best for different segments of the sequence of examples. We call such bounds shifting bounds. They hold for arbitrary sequences of examples and arbitrary sequences of predictors.Naturally shifting bounds are much harder to prove. The only known bounds are for the case when the comparison class consists of a sequences of experts or boolean disjunctions. In this paper we develop the methodology for lifting known static bounds to the shifting case. In particular we obtain bounds when the comparison class consists of linear neurons (linear combinations of experts). Our essential technique is to project the hypothesis of the static algorithm at the end of each trial into a suitably chosen convex region. This keeps the hypothesis of the algorithm well-behaved and the static bounds can be converted to shifting bounds.

234 citations

••

TL;DR: Algorithms that learn to improve search performance on large-scale optimization tasks, including STAGE, which works by learning an evaluation function that predicts the outcome of a local search algorithm from features of states visited during search.

Abstract: This paper describes algorithms that learn to improve search performance on large-scale optimization tasks. The main algorithm, STAGE, works by learning an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is then used to bias future search trajectories toward better optima on the same problem. Another algorithm, X-STAGE, transfers previously learned evaluation functions to new, similar optimization problems. Empirical results are provided on seven large-scale optimization domains: bin-packing, channel routing, Bayesian network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.

185 citations

••

TL;DR: An algorithm for finding principal manifolds that can be regularized in a variety of ways is proposed and bounds on the covering numbers are given which allows us to obtain nearly optimal learning rates for certain types of regularization operators.

Abstract: Many settings of unsupervised learning can be viewed as quantization problems - the minimization of the expected quantization error subject to some restrictions. This allows the use of tools such as regularization from the theory of (supervised) risk minimization for unsupervised learning. This setting turns out to be closely related to principal curves, the generative topographic map, and robust coding.We explore this connection in two ways: (1) we propose an algorithm for finding principal manifolds that can be regularized in a variety of ways; and (2) we derive uniform convergence bounds and hence bounds on the learning rates of the algorithm. In particular, we give bounds on the covering numbers which allows us to obtain nearly optimal learning rates for certain types of regularization operators. Experimental results demonstrate the feasibility of the approach.

114 citations

••

TL;DR: A family of gradient descent algorithms for learning linear functions in an online setting that includes the classical LMS algorithm as well as new variants such as the Exponentiated Gradient (EG) algorithm due to Kivinen and Warmuth are considered.

Abstract: A family of gradient descent algorithms for learning linear functions in an online setting is considered. The family includes the classical LMS algorithm as well as new variants such as the Exponentiated Gradient (EG) algorithm due to Kivinen and Warmuth. The algorithms are based on prior distributions defined on the weight space. Techniques from differential geometry are used to develop the algorithms as gradient descent iterations with respect to the natural gradient in the Riemannian structure induced by the prior distribution. The proposed framework subsumes the notion of "link-functions".