Header logo is ei


2008


no image
Frequent Subgraph Retrieval in Geometric Graph Databases

Nowozin, S., Tsuda, K.

(180), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, November 2008 (techreport)

Abstract
Discovery of knowledge from geometric graph databases is of particular importance in chemistry and biology, because chemical compounds and proteins are represented as graphs with 3D geometric coordinates. In such applications, scientists are not interested in the statistics of the whole database. Instead they need information about a novel drug candidate or protein at hand, represented as a query graph. We propose a polynomial-delay algorithm for geometric frequent subgraph retrieval. It enumerates all subgraphs of a single given query graph which are frequent geometric epsilon-subgraphs under the entire class of rigid geometric transformations in a database. By using geometric epsilon-subgraphs, we achieve tolerance against variations in geometry. We compare the proposed algorithm to gSpan on chemical compound data, and we show that for a given minimum support the total number of frequent patterns is substantially limited by requiring geometric matching. Although the computation time per pattern is larger than for non-geometric graph mining, the total time is within a reasonable level even for small minimum support.

PDF [BibTex]

2008

PDF [BibTex]


no image
Simultaneous Implicit Surface Reconstruction and Meshing

Giesen, J., Maier, M., Schölkopf, B.

(179), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, November 2008 (techreport)

Abstract
We investigate an implicit method to compute a piecewise linear representation of a surface from a set of sample points. As implicit surface functions we use the weighted sum of piecewise linear kernel functions. For such a function we can partition Rd in such a way that these functions are linear on the subsets of the partition. For each subset in the partition we can then compute the zero level set of the function exactly as the intersection of a hyperplane with the subset.

PDF [BibTex]

PDF [BibTex]


no image
Taxonomy Inference Using Kernel Dependence Measures

Blaschko, M., Gretton, A.

(181), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, November 2008 (techreport)

Abstract
We introduce a family of unsupervised algorithms, numerical taxonomy clustering, to simultaneously cluster data, and to learn a taxonomy that encodes the relationship between the clusters. The algorithms work by maximizing the dependence between the taxonomy and the original data. The resulting taxonomy is a more informative visualization of complex data than simple clustering; in addition, taking into account the relations between different clusters is shown to substantially improve the quality of the clustering, when compared with state-of-the-art algorithms in the literature (both spectral clustering and a previous dependence maximization approach). We demonstrate our algorithm on image and text data.

PDF [BibTex]

PDF [BibTex]


no image
Infinite Kernel Learning

Gehler, P., Nowozin, S.

(178), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, October 2008 (techreport)

Abstract
In this paper we consider the problem of automatically learning the kernel from general kernel classes. Specifically we build upon the Multiple Kernel Learning (MKL) framework and in particular on the work of (Argyriou, Hauser, Micchelli, & Pontil, 2006). We will formulate a Semi-Infinite Program (SIP) to solve the problem and devise a new algorithm to solve it (Infinite Kernel Learning, IKL). The IKL algorithm is applicable to both the finite and infinite case and we find it to be faster and more stable than SimpleMKL (Rakotomamonjy, Bach, Canu, & Grandvalet, 2007) for cases of many kernels. In the second part we present the first large scale comparison of SVMs to MKL on a variety of benchmark datasets, also comparing IKL. The results show two things: a) for many datasets there is no benefit in linearly combining kernels with MKL/IKL instead of the SVM classifier, thus the flexibility of using more than one kernel seems to be of no use, b) on some datasets IKL yields impressive increases in accuracy over SVM/MKL due to the possibility of using a largely increased kernel set. In those cases, IKL remains practical, whereas both cross-validation or standard MKL is infeasible.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Large Scale Variational Inference and Experimental Design for Sparse Generalized Linear Models

Seeger, M., Nickisch, H.

(175), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, September 2008 (techreport)

PDF [BibTex]

PDF [BibTex]


no image
Block-Iterative Algorithms for Non-Negative Matrix Approximation

Sra, S.

(176), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, September 2008 (techreport)

Abstract
In this report we present new algorithms for non-negative matrix approximation (NMA), commonly known as the NMF problem. Our methods improve upon the well-known methods of Lee & Seung [19] for both the Frobenius norm as well the Kullback-Leibler divergence versions of the problem. For the latter problem, our results are especially interesting because it seems to have witnessed much lesser algorithmic progress as compared to the Frobenius norm NMA problem. Our algorithms are based on a particular block-iterative acceleration technique for EM, which preserves the multiplicative nature of the updates and also ensures monotonicity. Furthermore, our algorithms also naturally apply to the Bregman-divergence NMA algorithms of Dhillon and Sra [8]. Experimentally, we show that our algorithms outperform the traditional Lee/Seung approach most of the time.

PDF [BibTex]

PDF [BibTex]


no image
Approximation Algorithms for Bregman Clustering Co-clustering and Tensor Clustering

Sra, S., Jegelka, S., Banerjee, A.

(177), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, September 2008 (techreport)

Abstract
The Euclidean K-means problem is fundamental to clustering and over the years it has been intensely investigated. More recently, generalizations such as Bregman k-means [8], co-clustering [10], and tensor (multi-way) clustering [40] have also gained prominence. A well-known computational difficulty encountered by these clustering problems is the NP-Hardness of the associated optimization task, and commonly used methods guarantee at most local optimality. Consequently, approximation algorithms of varying degrees of sophistication have been developed, though largely for the basic Euclidean K-means (or `1-norm K-median) problem. In this paper we present approximation algorithms for several Bregman clustering problems by building upon the recent paper of Arthur and Vassilvitskii [5]. Our algorithms obtain objective values within a factor O(logK) for Bregman k-means, Bregman co-clustering, Bregman tensor clustering, and weighted kernel k-means. To our knowledge, except for some special cases, approximation algorithms have not been considered for these general clustering problems. There are several important implications of our work: (i) under the same assumptions as Ackermann et al. [1] it yields a much faster algorithm (non-exponential in K, unlike [1]) for information-theoretic clustering, (ii) it answers several open problems posed by [4], including generalizations to Bregman co-clustering, and tensor clustering, (iii) it provides practical and easy to implement methods—in contrast to several other common approximation approaches.

PDF [BibTex]

PDF [BibTex]


no image
Combining Appearance and Motion for Human Action Classification in Videos

Dhillon, P., Nowozin, S., Lampert, C.

(174), Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany, August 2008 (techreport)

Abstract
We study the question of activity classification in videos and present a novel approach for recognizing human action categories in videos by combining information from appearance and motion of human body parts. Our approach uses a tracking step which involves Particle Filtering and a local non - parametric clustering step. The motion information is provided by the trajectory of the cluster modes of a local set of particles. The statistical information about the particles of that cluster over a number of frames provides the appearance information. Later we use a “Bag ofWords” model to build one histogram per video sequence from the set of these robust appearance and motion descriptors. These histograms provide us characteristic information which helps us to discriminate among various human actions and thus classify them correctly. We tested our approach on the standard KTH and Weizmann human action datasets and the results were comparable to the state of the art. Additionally our approach is able to distinguish between activities that involve the motion of complete body from those in which only certain body parts move. In other words, our method discriminates well between activities with “gross motion” like running, jogging etc. and “local motion” like waving, boxing etc.

PDF [BibTex]

PDF [BibTex]


no image
Example-based Learning for Single-image Super-resolution and JPEG Artifact Removal

Kim, K., Kwon, Y.

(173), Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany, August 2008 (techreport)

Abstract
This paper proposes a framework for single-image super-resolution and JPEG artifact removal. The underlying idea is to learn a map from input low-quality images (suitably preprocessed low-resolution or JPEG encoded images) to target high-quality images based on example pairs of input and output images. To retain the complexity of the resulting learning problem at a moderate level, a patch-based approach is taken such that kernel ridge regression (KRR) scans the input image with a small window (patch) and produces a patchvalued output for each output pixel location. These constitute a set of candidate images each of which reflects different local information. An image output is then obtained as a convex combination of candidates for each pixel based on estimated confidences of candidates. To reduce the time complexity of training and testing for KRR, a sparse solution is found by combining the ideas of kernel matching pursuit and gradient descent. As a regularized solution, KRR leads to a better generalization than simply storing the examples as it has been done in existing example-based super-resolution algorithms and results in much less noisy images. However, this may introduce blurring and ringing artifacts around major edges as sharp changes are penalized severely. A prior model of a generic image class which takes into account the discontinuity property of images is adopted to resolve this problem. Comparison with existing super-resolution and JPEG artifact removal methods shows the effectiveness of the proposed method. Furthermore, the proposed method is generic in that it has the potential to be applied to many other image enhancement applications.

PDF [BibTex]

PDF [BibTex]


no image
Unsupervised Bayesian Time-series Segmentation based on Linear Gaussian State-space Models

Chiappa, S.

(171), Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany, June 2008 (techreport)

Abstract
Unsupervised time-series segmentation in the general scenario in which the number of segment-types and segment boundaries are a priori unknown is a fundamental problem in many applications and requires an accurate segmentation model as well as a way of determining an appropriate number of segment-types. In most approaches, segmentation and determination of number of segment-types are addressed in two separate steps, since the segmentation model assumes a predefined number of segment-types. The determination of number of segment-types is thus achieved by training and comparing several separate models. In this paper, we take a Bayesian approach to a segmentation model based on linear Gaussian state-space models to achieve structure selection within the model. An appropriate prior distribution on the parameters is used to enforce a sparse parametrization, such that the model automatically selects the smallest number of underlying dynamical systems that explain the data well and a parsimonious structure for each dynamical system. As the resulting model is computationally intractable, we introduce a variational approximation, in which a reformulation of the problem enables to use an efficient inference algorithm.

[BibTex]

[BibTex]


no image
A New Non-monotonic Gradient Projection Method for the Non-negative Least Squares Problem

Kim, D., Sra, S., Dhillon, I.

(TR-08-28), University of Texas, Austin, TX, USA, June 2008 (techreport)

Web [BibTex]

Web [BibTex]


no image
Non-monotonic Poisson Likelihood Maximization

Sra, S., Kim, D., Schölkopf, B.

(170), Max-Planck Institute for Biological Cybernetics, Tübingen, Germany, June 2008 (techreport)

Abstract
This report summarizes the theory and some main applications of a new non-monotonic algorithm for maximizing a Poisson Likelihood, which for Positron Emission Tomography (PET) is equivalent to minimizing the associated Kullback-Leibler Divergence, and for Transmission Tomography is similar to maximizing the dual of a maximum entropy problem. We call our method non-monotonic maximum likelihood (NMML) and show its application to different problems such as tomography and image restoration. We discuss some theoretical properties such as convergence for our algorithm. Our experimental results indicate that speedups obtained via our non-monotonic methods are substantial.

PDF [BibTex]

PDF [BibTex]


no image
A Kernel Method for the Two-sample Problem

Gretton, A., Borgwardt, K., Rasch, M., Schölkopf, B., Smola, A.

(157), Max-Planck-Institute for Biological Cybernetics Tübingen, April 2008 (techreport)

Abstract
We propose a framework for analyzing and comparing distributions, allowing us to design statistical tests to determine if two samples are drawn from different distributions. Our test statistic is the largest difference in expectations over functions in the unit ball of a reproducing kernel Hilbert space (RKHS). We present two tests based on large deviation bounds for the test statistic, while a third is based on the asymptotic distribution of this statistic. The test statistic can be computed in quadratic time, although efficient linear time approximations are available. Several classical metrics on distributions are recovered when the function space used to compute the difference in expectations is allowed to be more general (eg.~a Banach space). We apply our two-sample tests to a variety of problems, including attribute matching for databases using the Hungarian marriage method, where they perform strongly. Excellent performance is also obtained when comparing distributions over graphs, for which these are the first such tests.

PDF [BibTex]

PDF [BibTex]


no image
Energy Functionals for Manifold-valued Mappings and Their Properties

Hein, M., Steinke, F., Schölkopf, B.

(167), Max Planck Institute for Biological Cybernetics, Tübingen, January 2008 (techreport)

Abstract
This technical report is merely an extended version of the appendix of Steinke et.al. "Manifold-valued Thin-Plate Splines with Applications in Computer Graphics" (2008) with complete proofs, which had to be omitted due to space restrictions. This technical report requires a basic knowledge of differential geometry. However, apart from that requirement the technical report is self-contained.

PDF [BibTex]

PDF [BibTex]

2007


no image
Bayesian Estimators for Robins-Ritov’s Problem

Harmeling, S., Toussaint, M.

(EDI-INF-RR-1189), School of Informatics, University of Edinburgh, October 2007 (techreport)

Abstract
Bayesian or likelihood-based approaches to data analysis became very popular in the field of Machine Learning. However, there exist theoretical results which question the general applicability of such approaches; among those a result by Robins and Ritov which introduce a specific example for which they prove that a likelihood-based estimator will fail (i.e. it does for certain cases not converge to a true parameter estimate, even given infinite data). In this paper we consider various approaches to formulate likelihood-based estimators in this example, basically by considering various extensions of the presumed generative model of the data. We can derive estimators which are very similar to the classical Horvitz-Thompson and which also account for a priori knowledge of an observation probability function.

PDF [BibTex]

2007

PDF [BibTex]


no image
Learning with Transformation Invariant Kernels

Walder, C., Chapelle, O.

(165), Max Planck Institute for Biological Cybernetics, Tübingen, Germany, September 2007 (techreport)

Abstract
Abstract. This paper considers kernels invariant to translation, rotation and dilation. We show that no non-trivial positive definite (p.d.) kernels exist which are radial and dilation invariant, only conditionally positive definite (c.p.d.) ones. Accordingly, we discuss the c.p.d. case and provide some novel analysis, including an elementary derivation of a c.p.d. representer theorem. On the practical side, we give a support vector machine (s.v.m.) algorithm for arbitrary c.p.d. kernels. For the thin-plate kernel this leads to a classifier with only one parameter (the amount of regularisation), which we demonstrate to be as effective as an s.v.m. with the Gaussian kernel, even though the Gaussian involves a second parameter (the length scale).

PDF [BibTex]

PDF [BibTex]


no image
Scalable Semidefinite Programming using Convex Perturbations

Kulis, B., Sra, S., Jegelka, S.

(TR-07-47), University of Texas, Austin, TX, USA, September 2007 (techreport)

Abstract
Several important machine learning problems can be modeled and solved via semidefinite programs. Often, researchers invoke off-the-shelf software for the associated optimization, which can be inappropriate for many applications due to computational and storage requirements. In this paper, we introduce the use of convex perturbations for semidefinite programs (SDPs). Using a particular perturbation function, we arrive at an algorithm for SDPs that has several advantages over existing techniques: a) it is simple, requiring only a few lines of MATLAB, b) it is a first-order method which makes it scalable, c) it can easily exploit the structure of a particular SDP to gain efficiency (e.g., when the constraint matrices are low-rank). We demonstrate on several machine learning applications that the proposed algorithm is effective in finding fast approximations to large-scale SDPs.

PDF [BibTex]

PDF [BibTex]


no image
Sparse Multiscale Gaussian Process Regression

Walder, C., Kim, K., Schölkopf, B.

(162), Max Planck Institute for Biological Cybernetics, Tübingen, Germany, August 2007 (techreport)

Abstract
Most existing sparse Gaussian process (g.p.) models seek computational advantages by basing their computations on a set of m basis functions that are the covariance function of the g.p. with one of its two inputs fixed. We generalise this for the case of Gaussian covariance function, by basing our computations on m Gaussian basis functions with arbitrary diagonal covariance matrices (or length scales). For a fixed number of basis functions and any given criteria, this additional flexibility permits approximations no worse and typically better than was previously possible. Although we focus on g.p. regression, the central idea is applicable to all kernel based algorithms, such as the support vector machine. We perform gradient based optimisation of the marginal likelihood, which costs O(m2n) time where n is the number of data points, and compare the method to various other sparse g.p. methods. Our approach outperforms the other methods, particularly for the case of very few basis functions, i.e. a very high sparsity ratio.

PDF [BibTex]

PDF [BibTex]


no image
Efficient Subwindow Search for Object Localization

Blaschko, M., Hofmann, T., Lampert, C.

(164), Max Planck Institute for Biological Cybernetics, Tübingen, Germany, August 2007 (techreport)

Abstract
Recent years have seen huge advances in object recognition from images. Recognition rates beyond 95% are the rule rather than the exception on many datasets. However, most state-of-the-art methods can only decide if an object is present or not. They are not able to provide information on the object location or extent within in the image. We report on a simple yet powerful scheme that extends many existing recognition methods to also perform localization of object bounding boxes. This is achieved by maximizing the classification score over all possible subrectangles in the image. Despite the impression that this would be computationally intractable, we show that in many situations efficient algorithms exist which solve a generalized maximum subrectangle problem. We show how our method is applicable to a variety object detection frameworks and demonstrate its performance by applying it to the popular bag of visual words model, achieving competitive results on the PASCAL VOC 2006 dataset.

PDF [BibTex]

PDF [BibTex]


no image
Cluster Identification in Nearest-Neighbor Graphs

Maier, M., Hein, M., von Luxburg, U.

(163), Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany, May 2007 (techreport)

Abstract
Assume we are given a sample of points from some underlying distribution which contains several distinct clusters. Our goal is to construct a neighborhood graph on the sample points such that clusters are ``identified‘‘: that is, the subgraph induced by points from the same cluster is connected, while subgraphs corresponding to different clusters are not connected to each other. We derive bounds on the probability that cluster identification is successful, and use them to predict ``optimal‘‘ values of k for the mutual and symmetric k-nearest-neighbor graphs. We point out different properties of the mutual and symmetric nearest-neighbor graphs related to the cluster identification problem.

PDF [BibTex]

PDF [BibTex]


no image
Exploring model selection techniques for nonlinear dimensionality reduction

Harmeling, S.

(EDI-INF-RR-0960), School of Informatics, University of Edinburgh, March 2007 (techreport)

Abstract
Nonlinear dimensionality reduction (NLDR) methods have become useful tools for practitioners who are faced with the analysis of high-dimensional data. Of course, not all NLDR methods are equally applicable to a particular dataset at hand. Thus it would be useful to come up with model selection criteria that help to choose among different NLDR algorithms. This paper explores various approaches to this problem and evaluates them on controlled data sets. Comprehensive experiments will show that model selection scores based on stability are not useful, while scores based on Gaussian processes are helpful for the NLDR problem.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Dirichlet Mixtures of Bayesian Linear Gaussian State-Space Models: a Variational Approach

Chiappa, S., Barber, D.

(161), Max Planck Institute for Biological Cybernetics, Tübingen, Germany, March 2007 (techreport)

Abstract
We describe two related models to cluster multidimensional time-series under the assumption of an underlying linear Gaussian dynamical process. In the first model, times-series are assigned to the same cluster when they show global similarity in their dynamics, while in the second model times-series are assigned to the same cluster when they show simultaneous similarity. Both models are based on Dirichlet Mixtures of Bayesian Linear Gaussian State-Space Models in order to (semi) automatically determine an appropriate number of components in the mixture, and to additionally bias the components to a parsimonious parameterization. The resulting models are formally intractable and to deal with this we describe a deterministic approximation based on a novel implementation of Variational Bayes.

PDF [BibTex]

PDF [BibTex]


no image
Modeling data using directional distributions: Part II

Sra, S., Jain, P., Dhillon, I.

(TR-07-05), University of Texas, Austin, TX, USA, February 2007 (techreport)

Abstract
High-dimensional data is central to most data mining applications, and only recently has it been modeled via directional distributions. In [Banerjee et al., 2003] the authors introduced the use of the von Mises-Fisher (vMF) distribution for modeling high-dimensional directional data, particularly for text and gene expression analysis. The vMF distribution is one of the simplest directional distributions. TheWatson, Bingham, and Fisher-Bingham distributions provide distri- butions with an increasing number of parameters and thereby commensurately increased modeling power. This report provides a followup study to the initial development in [Banerjee et al., 2003] by presenting Expectation Maximization (EM) procedures for estimating parameters of a mixture of Watson (moW) distributions. The numerical challenges associated with parameter estimation for both of these distributions are significantly more difficult than for the vMF distribution. We develop new numerical approximations for estimating the parameters permitting us to model real- life data more accurately. Our experimental results establish that for certain data sets improved modeling power translates into better results.

PDF [BibTex]

PDF [BibTex]


no image
Automatic 3D Face Reconstruction from Single Images or Video

Breuer, P., Kim, K., Kienzle, W., Blanz, V., Schölkopf, B.

(160), Max Planck Institute for Biological Cybernetics, Tübingen, Germany, February 2007 (techreport)

Abstract
This paper presents a fully automated algorithm for reconstructing a textured 3D model of a face from a single photograph or a raw video stream. The algorithm is based on a combination of Support Vector Machines (SVMs) and a Morphable Model of 3D faces. After SVM face detection, individual facial features are detected using a novel regression-and classification-based approach, and probabilistically plausible configurations of features are selected to produce a list of candidates for several facial feature positions. In the next step, the configurations of feature points are evaluated using a novel criterion that is based on a Morphable Model and a combination of linear projections. Finally, the feature points initialize a model-fitting procedure of the Morphable Model. The result is a high-resolution 3D surface model.

PDF [BibTex]

PDF [BibTex]


no image
Relative Entropy Policy Search

Peters, J.

CLMC Technical Report: TR-CLMC-2007-2, Computational Learning and Motor Control Lab, Los Angeles, CA, 2007, clmc (techreport)

Abstract
This technical report describes a cute idea of how to create new policy search approaches. It directly relates to the Natural Actor-Critic methods but allows the derivation of one shot solutions. Future work may include the application to interesting problems.

PDF link (url) [BibTex]

PDF link (url) [BibTex]

2001


no image
Kernel Methods for Extracting Local Image Semantics

Bradshaw, B., Schölkopf, B., Platt, J.

(MSR-TR-2001-99), Microsoft Research, October 2001 (techreport)

Web [BibTex]

2001

Web [BibTex]


no image
Calibration of Digital Amateur Cameras

Urbanek, M., Horaud, R., Sturm, P.

(RR-4214), INRIA Rhone Alpes, Montbonnot, France, July 2001 (techreport)

Web [BibTex]

Web [BibTex]


no image
Incorporating Invariances in Non-Linear Support Vector Machines

Chapelle, O., Schölkopf, B.

Max Planck Institute for Biological Cybernetics / Biowulf Technologies, 2001 (techreport)

Abstract
We consider the problem of how to incorporate in the Support Vector Machine (SVM) framework invariances given by some a priori known transformations under which the data should be invariant. It extends some previous work which was only applicable with linear SVMs and we show on a digit recognition task that the proposed approach is superior to the traditional Virtual Support Vector method.

PostScript [BibTex]

PostScript [BibTex]


no image
Bound on the Leave-One-Out Error for Density Support Estimation using nu-SVMs

Gretton, A., Herbrich, R., Schölkopf, B., Smola, A., Rayner, P.

University of Cambridge, 2001 (techreport)

[BibTex]

[BibTex]


no image
Bound on the Leave-One-Out Error for 2-Class Classification using nu-SVMs

Gretton, A., Herbrich, R., Schölkopf, B., Rayner, P.

University of Cambridge, 2001, Updated May 2003 (literature review expanded) (techreport)

Abstract
Three estimates of the leave-one-out error for $nu$-support vector (SV) machine binary classifiers are presented. Two of the estimates are based on the geometrical concept of the {em span}, which was introduced in the context of bounding the leave-one-out error for $C$-SV machine binary classifiers, while the third is based on optimisation over the criterion used to train the $nu$-support vector classifier. It is shown that the estimates presented herein provide informative and efficient approximations of the generalisation behaviour, in both a toy example and benchmark data sets. The proof strategies in the $nu$-SV context are also compared with those used to derive leave-one-out error estimates in the $C$-SV case.

PostScript [BibTex]

PostScript [BibTex]


no image
Some kernels for structured data

Bartlett, P., Schölkopf, B.

Biowulf Technologies, 2001 (techreport)

[BibTex]

[BibTex]


no image
Inference Principles and Model Selection

Buhmann, J., Schölkopf, B.

(01301), Dagstuhl Seminar, 2001 (techreport)

Web [BibTex]

Web [BibTex]