Header logo is ei


2006


no image
Large Scale Multiple Kernel Learning

Sonnenburg, S., Rätsch, G., Schäfer, C., Schölkopf, B.

Journal of Machine Learning Research, 7, pages: 1531-1565, July 2006 (article)

Abstract
While classical kernel-based learning algorithms are based on a single kernel, in practice it is often desirable to use multiple kernels. Lanckriet et al. (2004) considered conic combinations of kernel matrices for classification, leading to a convex quadratically constrained quadratic program. We show that it can be rewritten as a semi-infinite linear program that can be efficiently solved by recycling the standard SVM implementations. Moreover, we generalize the formulation and our method to a larger class of problems, including regression and one-class classification. Experimental results show that the proposed algorithm works for hundred thousands of examples or hundreds of kernels to be combined, and helps for automatic model selection, improving the interpretability of the learning result. In a second part we discuss general speed up mechanism for SVMs, especially when used with sparse feature maps as appear for string kernels, allowing us to train a string kernel SVM on a 10 million real-world splice data set from computational biology. We integrated multiple kernel learning in our machine learning toolbox SHOGUN for which the source code is publicly available at http://www.fml.tuebingen.mpg.de/raetsch/projects/shogun.

PDF [BibTex]

2006

PDF [BibTex]


no image
Factorial coding of natural images: how effective are linear models in removing higher-order dependencies?

Bethge, M.

Journal of the Optical Society of America A, 23(6):1253-1268, June 2006 (article)

Abstract
The performance of unsupervised learning models for natural images is evaluated quantitatively by means of information theory. We estimate the gain in statistical independence (the multi-information reduction) achieved with independent component analysis (ICA), principal component analysis (PCA), zero-phase whitening, and predictive coding. Predictive coding is translated into the transform coding framework, where it can be characterized by the constraint of a triangular filter matrix. A randomly sampled whitening basis and the Haar wavelet are included into the comparison as well. The comparison of all these methods is carried out for different patch sizes, ranging from 2x2 to 16x16 pixels. In spite of large differences in the shape of the basis functions, we find only small differences in the multi-information between all decorrelation transforms (5% or less) for all patch sizes. Among the second-order methods, PCA is optimal for small patch sizes and predictive coding performs best for large patch sizes. The extra gain achieved with ICA is always less than 2%. In conclusion, the `edge filters‘ found with ICA lead only to a surprisingly small improvement in terms of its actual objective.

PDF Web [BibTex]


no image
Classifying EEG and ECoG Signals without Subject Training for Fast BCI Implementation: Comparison of Non-Paralysed and Completely Paralysed Subjects

Hill, N., Lal, T., Schröder, M., Hinterberger, T., Wilhelm, B., Nijboer, F., Mochty, U., Widman, G., Elger, C., Schölkopf, B., Kübler, A., Birbaumer, N.

IEEE Transactions on Neural Systems and Rehabilitation Engineering, 14(2):183-186, June 2006 (article)

Abstract
We summarize results from a series of related studies that aim to develop a motor-imagery-based brain-computer interface using a single recording session of EEG or ECoG signals for each subject. We apply the same experimental and analytical methods to 11 non-paralysed subjects (8 EEG, 3 ECoG), and to 5 paralysed subjects (4 EEG, 1 ECoG) who had been unable to communicate for some time. While it was relatively easy to obtain classifiable signals quickly from most of the non-paralysed subjects, it proved impossible to classify the signals obtained from the paralysed patients by the same methods. This highlights the fact that though certain BCI paradigms may work well with healthy subjects, this does not necessarily indicate success with the target user group. We outline possible reasons for this failure to transfer.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
SCARNA: Fast and Accurate Structural Alignment of RNA Sequences by Matching Fixed-Length Stem Fragments

Tabei, Y., Tsuda, K., Kin, T., Asai, K.

Bioinformatics, 22(14):1723-1729, May 2006 (article)

Abstract
The functions of non-coding RNAs are strongly related to their secondary structures, but it is known that a secondary structure prediction of a single sequence is not reliable. Therefore, we have to collect similar RNA sequences with a common secondary structure for the analyses of a new non-coding RNA without knowing the exact secondary structure itself. Therefore, the sequence comparison in searching similar RNAs should consider not only their sequence similarities but their potential secondary structures. Sankoff‘s algorithm predicts the common secondary structures of the sequences, but it is computationally too expensive to apply to large-scale analyses. Because we often want to compare a large number of cDNA sequences or to search similar RNAs in the whole genome sequences, much faster algorithms are required. We propose a new method of comparing RNA sequences based on the structural alignments of the fixed-length fragments of the stem candidates. The implemented software, SCARNA (Stem Candidate Aligner for RNAs), is fast enough to apply to the long sequences in the large-scale analyses. The accuracy of the alignments is better or comparable to the much slower existing algorithms.

PDF Web DOI [BibTex]


no image
The Effect of Artifacts on Dependence Measurement in fMRI

Gretton, A., Belitski, A., Murayama, Y., Schölkopf, B., Logothetis, N.

Magnetic Resonance Imaging, 24(4):401-409, April 2006 (article)

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Phase noise and the classification of natural images

Wichmann, F., Braun, D., Gegenfurtner, K.

Vision Research, 46(8-9):1520-1529, April 2006 (article)

Abstract
We measured the effect of global phase manipulations on a rapid animal categorization task. The Fourier spectra of our images of natural scenes were manipulated by adding zero-mean random phase noise at all spatial frequencies. The phase noise was the independent variable, uniformly and symmetrically distributed between 0 degree and ±180 degrees. Subjects were remarkably resistant to phase noise. Even with ±120 degree phase noise subjects were still performing at 75% correct. The high resistance of the subjects’ animal categorization rate to phase noise suggests that the visual system is highly robust to such random image changes. The proportion of correct answers closely followed the correlation between original and the phase noise-distorted images. Animal detection rate was higher when the same task was performed with contrast reduced versions of the same natural images, at contrasts where the contrast reduction mimicked that resulting from our phase randomization. Since the subjects’ categorization rate was better in the contrast experiment, reduction of local contrast alone cannot explain the performance in the phase noise experiment. This result obtained with natural images differs from those obtained for simple sinusoidal stimuli were performance changes due to phase changes are attributed to local contrast changes only. Thus the global phasechange accompanying disruption of image structure such as edges and object boundaries at different spatial scales reduces object classification over and above the performance deficit resulting from reducing contrast. Additional colour information improves the categorization performance by 2 %.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
A Direct Method for Building Sparse Kernel Learning Algorithms

Wu, M., Schölkopf, B., BakIr, G.

Journal of Machine Learning Research, 7, pages: 603-624, April 2006 (article)

Abstract
Many Kernel Learning Algorithms(KLA), including Support Vector Machine (SVM), result in a Kernel Machine (KM), such as a kernel classifier, whose key component is a weight vector in a feature space implicitly introduced by a positive definite kernel function. This weight vector is usually obtained by solving a convex optimization problem. Based on this fact we present a direct method to build Sparse Kernel Learning Algorithms (SKLA) by adding one more constraint to the original convex optimization problem, such that the sparseness of the resulting KM is explicitly controlled while at the same time the performance of the resulting KM can be kept as high as possible. A gradient based approach is provided to solve this modified optimization problem. Applying this method to the SVM results in a concrete algorithm for building Sparse Large Margin Classifiers (SLMC). Further analysis of the SLMC algorithm indicates that it essentially finds a discriminating subspace that can be spanned by a small number of vectors, and in this subspace, the different classes of data are linearly well separated. Experimental results over several classification benchmarks demonstrate the effectiveness of our approach.

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
Statistical Properties of Kernel Principal Component Analysis

Blanchard, G., Bousquet, O., Zwald, L.

Machine Learning, 66(2-3):259-294, March 2006 (article)

Abstract
We study the properties of the eigenvalues of Gram matrices in a non-asymptotic setting. Using local Rademacher averages, we provide data-dependent and tight bounds for their convergence towards eigenvalues of the corresponding kernel operator. We perform these computations in a functional analytic framework which allows to deal implicitly with reproducing kernel Hilbert spaces of infinite dimension. This can have applications to various kernel algorithms, such as Support Vector Machines (SVM). We focus on Kernel Principal Component Analysis (KPCA) and, using such techniques, we obtain sharp excess risk bounds for the reconstruction error. In these bounds, the dependence on the decay of the spectrum and on the closeness of successive eigenvalues is made explicit.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Network-based de-noising improves prediction from microarray data

Kato, T., Murata, Y., Miura, K., Asai, K., Horton, P., Tsuda, K., Fujibuchi, W.

BMC Bioinformatics, 7(Suppl. 1):S4-S4, March 2006 (article)

Abstract
Prediction of human cell response to anti-cancer drugs (compounds) from microarray data is a challenging problem, due to the noise properties of microarrays as well as the high variance of living cell responses to drugs. Hence there is a strong need for more practical and robust methods than standard methods for real-value prediction. We devised an extended version of the off-subspace noise-reduction (de-noising) method to incorporate heterogeneous network data such as sequence similarity or protein-protein interactions into a single framework. Using that method, we first de-noise the gene expression data for training and test data and also the drug-response data for training data. Then we predict the unknown responses of each drug from the de-noised input data. For ascertaining whether de-noising improves prediction or not, we carry out 12-fold cross-validation for assessment of the prediction performance. We use the Pearson‘s correlation coefficient between the true and predicted respon se values as the prediction performance. De-noising improves the prediction performance for 65% of drugs. Furthermore, we found that this noise reduction method is robust and effective even when a large amount of artificial noise is added to the input data. We found that our extended off-subspace noise-reduction method combining heterogeneous biological data is successful and quite useful to improve prediction of human cell cancer drug responses from microarray data.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Model-based Design Analysis and Yield Optimization

Pfingsten, T., Herrmann, D., Rasmussen, C.

IEEE Transactions on Semiconductor Manufacturing, 19(4):475-486, February 2006 (article)

Abstract
Fluctuations are inherent to any fabrication process. Integrated circuits and micro-electro-mechanical systems are particularly affected by these variations, and due to high quality requirements the effect on the devices’ performance has to be understood quantitatively. In recent years it has become possible to model the performance of such complex systems on the basis of design specifications, and model-based Sensitivity Analysis has made its way into industrial engineering. We show how an efficient Bayesian approach, using a Gaussian process prior, can replace the commonly used brute-force Monte Carlo scheme, making it possible to apply the analysis to computationally costly models. We introduce a number of global, statistically justified sensitivity measures for design analysis and optimization. Two models of integrated systems serve us as case studies to introduce the analysis and to assess its convergence properties. We show that the Bayesian Monte Carlo scheme can save costly simulation runs and can ensure a reliable accuracy of the analysis.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Weighting of experimental evidence in macromolecular structure determination

Habeck, M., Rieping, W., Nilges, M.

Proceedings of the National Academy of Sciences of the United States of America, 103(6):1756-1761, February 2006 (article)

Abstract
The determination of macromolecular structures requires weighting of experimental evidence relative to prior physical information. Although it can critically affect the quality of the calculated structures, experimental data are routinely weighted on an empirical basis. At present, cross-validation is the most rigorous method to determine the best weight. We describe a general method to adaptively weight experimental data in the course of structure calculation. It is further shown that the necessity to define weights for the data can be completely alleviated. We demonstrate the method on a structure calculation from NMR data and find that the resulting structures are optimal in terms of accuracy and structural quality. Our method is devoid of the bias imposed by an empirical choice of the weight and has some advantages over estimating the weight by cross-validation.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Classification of Faces in Man and Machine

Graf, A., Wichmann, F., Bülthoff, H., Schölkopf, B.

Neural Computation, 18(1):143-165, January 2006 (article)

PDF Web [BibTex]

PDF Web [BibTex]

2003


no image
Concentration Inequalities for Sub-Additive Functions Using the Entropy Method

Bousquet, O.

Stochastic Inequalities and Applications, 56, pages: 213-247, Progress in Probability, (Editors: Giné, E., C. Houdré and D. Nualart), November 2003 (article)

Abstract
We obtain exponential concentration inequalities for sub-additive functions of independent random variables under weak conditions on the increments of those functions, like the existence of exponential moments for these increments. As a consequence of these general inequalities, we obtain refinements of Talagrand's inequality for empirical processes and new bounds for randomized empirical processes. These results are obtained by further developing the entropy method introduced by Ledoux.

PostScript [BibTex]

2003

PostScript [BibTex]


no image
Statistical Learning Theory, Capacity and Complexity

Schölkopf, B.

Complexity, 8(4):87-94, July 2003 (article)

Abstract
We give an exposition of the ideas of statistical learning theory, followed by a discussion of how a reinterpretation of the insights of learning theory could potentially also benefit our understanding of a certain notion of complexity.

Web DOI [BibTex]


no image
Dealing with large Diagonals in Kernel Matrices

Weston, J., Schölkopf, B., Eskin, E., Leslie, C., Noble, W.

Annals of the Institute of Statistical Mathematics, 55(2):391-408, June 2003 (article)

Abstract
In kernel methods, all the information about the training data is contained in the Gram matrix. If this matrix has large diagonal values, which arises for many types of kernels, then kernel methods do not perform well: We propose and test several methods for dealing with this problem by reducing the dynamic range of the matrix while preserving the positive definiteness of the Hessian of the quadratic programming problem that one has to solve when training a Support Vector Machine, which is a common kernel approach for pattern recognition.

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
The em Algorithm for Kernel Matrix Completion with Auxiliary Data

Tsuda, K., Akaho, S., Asai, K.

Journal of Machine Learning Research, 4, pages: 67-81, May 2003 (article)

PDF [BibTex]

PDF [BibTex]


no image
Constructing Descriptive and Discriminative Non-linear Features: Rayleigh Coefficients in Kernel Feature Spaces

Mika, S., Rätsch, G., Weston, J., Schölkopf, B., Smola, A., Müller, K.

IEEE Transactions on Pattern Analysis and Machine Intelligence, 25(5):623-628, May 2003 (article)

Abstract
We incorporate prior knowledge to construct nonlinear algorithms for invariant feature extraction and discrimination. Employing a unified framework in terms of a nonlinearized variant of the Rayleigh coefficient, we propose nonlinear generalizations of Fisher‘s discriminant and oriented PCA using support vector kernel functions. Extensive simulations show the utility of our approach.

DOI [BibTex]

DOI [BibTex]


no image
Tractable Inference for Probabilistic Data Models

Csato, L., Opper, M., Winther, O.

Complexity, 8(4):64-68, April 2003 (article)

Abstract
We present an approximation technique for probabilistic data models with a large number of hidden variables, based on ideas from statistical physics. We give examples for two nontrivial applications. © 2003 Wiley Periodicals, Inc.

PDF GZIP Web [BibTex]

PDF GZIP Web [BibTex]


no image
Feature selection and transduction for prediction of molecular bioactivity for drug design

Weston, J., Perez-Cruz, F., Bousquet, O., Chapelle, O., Elisseeff, A., Schölkopf, B.

Bioinformatics, 19(6):764-771, April 2003 (article)

Abstract
Motivation: In drug discovery a key task is to identify characteristics that separate active (binding) compounds from inactive (non-binding) ones. An automated prediction system can help reduce resources necessary to carry out this task. Results: Two methods for prediction of molecular bioactivity for drug design are introduced and shown to perform well in a data set previously studied as part of the KDD (Knowledge Discovery and Data Mining) Cup 2001. The data is characterized by very few positive examples, a very large number of features (describing three-dimensional properties of the molecules) and rather different distributions between training and test data. Two techniques are introduced specifically to tackle these problems: a feature selection method for unbalanced data and a classifier which adapts to the distribution of the the unlabeled test data (a so-called transductive method). We show both techniques improve identification performance and in conjunction provide an improvement over using only one of the techniques. Our results suggest the importance of taking into account the characteristics in this data which may also be relevant in other problems of a similar type.

Web [BibTex]


no image
Use of the Zero-Norm with Linear Models and Kernel Methods

Weston, J., Elisseeff, A., Schölkopf, B., Tipping, M.

Journal of Machine Learning Research, 3, pages: 1439-1461, March 2003 (article)

Abstract
We explore the use of the so-called zero-norm of the parameters of linear models in learning. Minimization of such a quantity has many uses in a machine learning context: for variable or feature selection, minimizing training error and ensuring sparsity in solutions. We derive a simple but practical method for achieving these goals and discuss its relationship to existing techniques of minimizing the zero-norm. The method boils down to implementing a simple modification of vanilla SVM, namely via an iterative multiplicative rescaling of the training data. Applications we investigate which aid our discussion include variable and feature selection on biological microarray data, and multicategory classification.

PDF PostScript PDF [BibTex]

PDF PostScript PDF [BibTex]


no image
An Introduction to Variable and Feature Selection.

Guyon, I., Elisseeff, A.

Journal of Machine Learning, 3, pages: 1157-1182, 2003 (article)

[BibTex]

[BibTex]


no image
New Approaches to Statistical Learning Theory

Bousquet, O.

Annals of the Institute of Statistical Mathematics, 55(2):371-389, 2003 (article)

Abstract
We present new tools from probability theory that can be applied to the analysis of learning algorithms. These tools allow to derive new bounds on the generalization performance of learning algorithms and to propose alternative measures of the complexity of the learning task, which in turn can be used to derive new learning algorithms.

PostScript [BibTex]

PostScript [BibTex]