Header logo is ei


2008


no image
Generalization and Similarity in Exemplar Models of Categorization: Insights from Machine Learning

Jäkel, F., Schölkopf, B., Wichmann, F.

Psychonomic Bulletin and Review, 15(2):256-271, April 2008 (article)

Abstract
Exemplar theories of categorization depend on similarity for explaining subjects’ ability to generalize to new stimuli. A major criticism of exemplar theories concerns their lack of abstraction mechanisms and thus, seemingly, generalization ability. Here, we use insights from machine learning to demonstrate that exemplar models can actually generalize very well. Kernel methods in machine learning are akin to exemplar models and very successful in real-world applications. Their generalization performance depends crucially on the chosen similaritymeasure. While similarity plays an important role in describing generalization behavior it is not the only factor that controls generalization performance. In machine learning, kernel methods are often combined with regularization techniques to ensure good generalization. These same techniques are easily incorporated in exemplar models. We show that the Generalized Context Model (Nosofsky, 1986) and ALCOVE (Kruschke, 1992) are closely related to a statistical model called kernel logistic regression. We argue that generalization is central to the enterprise of understanding categorization behavior and suggest how insights from machine learning can offer some guidance. Keywords: kernel, similarity, regularization, generalization, categorization.

PDF Web DOI [BibTex]


no image
Bayesian methods for protein structure determination

Habeck, M.

Machine Learning in Structural Bioinformatics, April 2008 (talk)

Web [BibTex]

Web [BibTex]


no image
Manifold-valued Thin-plate Splines with Applications in Computer Graphics

Steinke, F., Hein, M., Peters, J., Schölkopf, B.

Computer Graphics Forum, 27(2):437-448, April 2008 (article)

Abstract
We present a generalization of thin-plate splines for interpolation and approximation of manifold-valued data, and demonstrate its usefulness in computer graphics with several applications from different fields. The cornerstone of our theoretical framework is an energy functional for mappings between two Riemannian manifolds which is independent of parametrization and respects the geometry of both manifolds. If the manifolds are Euclidean, the energy functional reduces to the classical thin-plate spline energy. We show how the resulting optimization problems can be solved efficiently in many cases. Our example applications range from orientation interpolation and motion planning in animation over geometric modelling tasks to color interpolation.

PDF AVI Web DOI [BibTex]


no image
The Metric Nearness Problem

Brickell, J., Dhillon, I., Sra, S., Tropp, J.

SIAM Journal on Matrix Analysis and Applications, 30(1):375-396, April 2008 (article)

Abstract
Metric nearness refers to the problem of optimally restoring metric properties to distance measurements that happen to be nonmetric due to measurement errors or otherwise. Metric data can be important in various settings, for example, in clustering, classification, metric-based indexing, query processing, and graph theoretic approximation algorithms. This paper formulates and solves the metric nearness problem: Given a set of pairwise dissimilarities, find a “nearest” set of distances that satisfy the properties of a metric—principally the triangle inequality. For solving this problem, the paper develops efficient triangle fixing algorithms that are based on an iterative projection method. An intriguing aspect of the metric nearness problem is that a special case turns out to be equivalent to the all pairs shortest paths problem. The paper exploits this equivalence and develops a new algorithm for the latter problem using a primal-dual method. Applications to graph clustering are provided as an illustratio n. We include experiments that demonstrate the computational superiority of triangle fixing over general purpose convex programming software. Finally, we conclude by suggesting various useful extensions and generalizations to metric nearness.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Bayesian Inference and Optimal Design for the Sparse Linear Model

Seeger, MW.

Journal of Machine Learning Research, 9, pages: 759-813, April 2008 (article)

Abstract
The linear model with sparsity-favouring prior on the coefficients has important applications in many different domains. In machine learning, most methods to date search for maximum a posteriori sparse solutions and neglect to represent posterior uncertainties. In this paper, we address problems of Bayesian optimal design (or experiment planning), for which accurate estimates of uncertainty are essential. To this end, we employ expectation propagation approximate inference for the linear model with Laplace prior, giving new insight into numerical stability properties and proposing a robust algorithm. We also show how to estimate model hyperparameters by empirical Bayesian maximisation of the marginal likelihood, and propose ideas in order to scale up the method to very large underdetermined problems. We demonstrate the versatility of our framework on the application of gene regulatory network identification from micro-array expression data, where both the Laplace prior and the active experimental design approach are shown to result in significant improvements. We also address the problem of sparse coding of natural images, and show how our framework can be used for compressive sensing tasks.

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
Consistency of Spectral Clustering

von Luxburg, U., Belkin, M., Bousquet, O.

Annals of Statistics, 36(2):555-586, April 2008 (article)

Abstract
Consistency is a key property of statistical algorithms when the data is drawn from some underlying probability distribution. Surprisingly, despite decades of work, little is known about consistency of most clustering algorithms. In this paper we investigate consistency of the popular family of spectral clustering algorithms, which clusters the data with the help of eigenvectors of graph Laplacian matrices. We develop new methods to establish that for increasing sample size, those eigenvectors converge to the eigenvectors of certain limit operators. As a result we can prove that one of the two major classes of spectral clustering (normalized clustering) converges under very general conditions, while the other (unnormalized clustering) is only consistent under strong additional assumptions, which are not always satisfied in real data. We conclude that our analysis provides strong evidence for the superiority of normalized spectral clustering.

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Plant Classification from Bat-Like Echolocation Signals

Yovel, Y., Franz, MO., Stilz, P., Schnitzler, H-U.

PLoS Computational Biology, 4(3, e1000032):1-13, March 2008 (article)

Abstract
Classification of plants according to their echoes is an elementary component of bat behavior that plays an important role in spatial orientation and food acquisition. Vegetation echoes are, however, highly complex stochastic signals: from an acoustical point of view, a plant can be thought of as a three-dimensional array of leaves reflecting the emitted bat call. The received echo is therefore a superposition of many reflections. In this work we suggest that the classification of these echoes might not be such a troublesome routine for bats as formerly thought. We present a rather simple approach to classifying signals from a large database of plant echoes that were created by ensonifying plants with a frequency-modulated bat-like ultrasonic pulse. Our algorithm uses the spectrogram of a single echo from which it only uses features that are undoubtedly accessible to bats. We used a standard machine learning algorithm (SVM) to automatically extract suitable linear combinations of time and frequency cues from the spectrograms such that classification with high accuracy is enabled. This demonstrates that ultrasonic echoes are highly informative about the species membership of an ensonified plant, and that this information can be extracted with rather simple, biologically plausible analysis. Thus, our findings provide a new explanatory basis for the poorly understood observed abilities of bats in classifying vegetation and other complex objects.

Web DOI [BibTex]

Web DOI [BibTex]


no image
The role of stimulus correlations for population decoding in the retina

Schwartz, G., Macke, J., Berry, M.

Computational and Systems Neuroscience 2008 (COSYNE 2008), 5, pages: 172, March 2008 (poster)

PDF Web [BibTex]

PDF Web [BibTex]


no image
Causal Reasoning by Evaluating the Complexity of Conditional Densities with Kernel Methods

Sun, X., Janzing, D., Schölkopf, B.

Neurocomputing, 71(7-9):1248-1256, March 2008 (article)

Abstract
We propose a method to quantify the complexity of conditional probability measures by a Hilbert space seminorm of the logarithm of its density. The concept of reproducing kernel Hilbert spaces (RKHSs) is a flexible tool to define such a seminorm by choosing an appropriate kernel. We present several examples with artificial data sets where our kernel-based complexity measure is consistent with our intuitive understanding of complexity of densities. The intention behind the complexity measure is to provide a new approach to inferring causal directions. The idea is that the factorization of the joint probability measure P(effect, cause) into P(effect|cause)P(cause) leads typically to "simpler" and "smoother" terms than the factorization into P(cause|effect)P(effect). Since the conventional constraint-based approach of causal discovery is not able to determine the causal direction between only two variables, our inference principle can in particular be useful when combined with other existing methods. We provide several simple examples with real-world data where the true causal directions indeed lead to simpler (conditional) densities.

Web DOI [BibTex]


no image
Natural Actor-Critic

Peters, J., Schaal, S.

Neurocomputing, 71(7-9):1180-1190, March 2008 (article)

Abstract
In this paper, we suggest a novel reinforcement learning architecture, the Natural Actor-Critic. The actor updates are achieved using stochastic policy gradients em- ploying Amari’s natural gradient approach, while the critic obtains both the natural policy gradient and additional parameters of a value function simultaneously by lin- ear regression. We show that actor improvements with natural policy gradients are particularly appealing as these are independent of coordinate frame of the chosen policy representation, and can be estimated more efficiently than regular policy gra- dients. The critic makes use of a special basis function parameterization motivated by the policy-gradient compatible function approximation. We show that several well-known reinforcement learning methods such as the original Actor-Critic and Bradtke’s Linear Quadratic Q-Learning are in fact Natural Actor-Critic algorithms. Empirical evaluations illustrate the effectiveness of our techniques in comparison to previous methods, and also demonstrate their applicability for learning control on an anthropomorphic robot arm.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Inferring Spike Trains From Local Field Potentials

Rasch, M., Gretton, A., Murayama, Y., Maass, W., Logothetis, N.

Journal of Neurophysiology, 99(3):1461-1476, March 2008 (article)

Abstract
We investigated whether it is possible to infer spike trains solely on the basis of the underlying local field potentials (LFPs). Using support vector machines and linear regression models, we found that in the primary visual cortex (V1) of monkeys, spikes can indeed be inferred from LFPs, at least with moderate success. Although there is a considerable degree of variation across electrodes, the low-frequency structure in spike trains (in the 100-ms range) can be inferred with reasonable accuracy, whereas exact spike positions are not reliably predicted. Two kinds of features of the LFP are exploited for prediction: the frequency power of bands in the high gamma-range (40–90 Hz) and information contained in lowfrequency oscillations ( 10 Hz), where both phase and power modulations are informative. Information analysis revealed that both features code (mainly) independent aspects of the spike-to-LFP relationship, with the low-frequency LFP phase coding for temporally clustered spiking activity. Although both features and prediction quality are similar during seminatural movie stimuli and spontaneous activity, prediction performance during spontaneous activity degrades much more slowly with increasing electrode distance. The general trend of data obtained with anesthetized animals is qualitatively mirrored in that of a more limited data set recorded in V1 of non-anesthetized monkeys. In contrast to the cortical field potentials, thalamic LFPs (e.g., LFPs derived from recordings in the dorsal lateral geniculate nucleus) hold no useful information for predicting spiking activity.

Web DOI [BibTex]

Web DOI [BibTex]


no image
ISD: A Software Package for Bayesian NMR Structure Calculation

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

Bioinformatics, 24(8):1104-1105, February 2008 (article)

Abstract
SUMMARY: The conventional approach to calculating biomolecular structures from nuclear magnetic resonance (NMR) data is often viewed as subjective due to its dependence on rules of thumb for deriving geometric constraints and suitable values for theory parameters from noisy experimental data. As a result, it can be difficult to judge the precision of an NMR structure in an objective manner. The Inferential Structure Determination (ISD) framework, which has been introduced recently, addresses this problem by using Bayesian inference to derive a probability distribution that represents both the unknown structure and its uncertainty. It also determines additional unknowns, such as theory parameters, that normally need be chosen empirically. Here we give an overview of the ISD software package, which implements this methodology. AVAILABILITY: The program is available at http://www.bioc.cam.ac.uk/isd

Web DOI [BibTex]

Web DOI [BibTex]


no image
Probabilistic Structure Calculation

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

Comptes Rendus Chimie, 11(4-5):356-369, February 2008 (article)

Abstract
Molecular structures are usually calculated from experimental data with some method of energy minimisation or non-linear optimisation. Key aims of a structure calculation are to estimate the coordinate uncertainty, and to provide a meaningful measure of the quality of the fit to the data. We discuss approaches to optimally combine prior information and experimental data and the connection to probability theory. We analyse the appropriate statistics for NOEs and NOE-derived distances, and the related question of restraint potentials. Finally, we will discuss approaches to determine the appropriate weight on the experimental evidence and to obtain in this way an estimate of the data quality from the structure calculation. Whereas objective estimates of coordinates and their uncertainties can only be obtained by a full Bayesian treatment of the problem, standard structure calculation methods continue to play an important role. To obtain the full benefit of these methods, they should be founded on a rigorous Baye sian analysis.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Fast Projection-based Methods for the Least Squares Nonnegative Matrix Approximation Problem

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

Statistical Analysis and Data Mining, 1(1):38-51, February 2008 (article)

Abstract
Nonnegative matrix approximation (NNMA) is a popular matrix decomposition technique that has proven to be useful across a diverse variety of fields with applications ranging from document analysis and image processing to bioinformatics and signal processing. Over the years, several algorithms for NNMA have been proposed, e.g. Lee and Seung‘s multiplicative updates, alternating least squares (ALS), and gradient descent-based procedures. However, most of these procedures suffer from either slow convergence, numerical instability, or at worst, serious theoretical drawbacks. In this paper, we develop a new and improved algorithmic framework for the least-squares NNMA problem, which is not only theoretically well-founded, but also overcomes many deficiencies of other methods. Our framework readily admits powerful optimization techniques and as concrete realizations we present implementations based on the Newton, BFGS and conjugate gradient methods. Our algorithms provide numerical resu lts supe rior to both Lee and Seung‘s method as well as to the alternating least squares heuristic, which was reported to work well in some situations but has no theoretical guarantees[1]. Our approach extends naturally to include regularization and box-constraints without sacrificing convergence guarantees. We present experimental results on both synthetic and real-world datasets that demonstrate the superiority of our methods, both in terms of better approximations as well as computational efficiency.

Web DOI [BibTex]

Web DOI [BibTex]


no image
A Unifying Probabilistic Framework for Analyzing Residual Dipolar Couplings

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

Journal of Biomolecular NMR, 40(2):135-144, February 2008 (article)

Abstract
Residual dipolar couplings provide complementary information to the nuclear Overhauser effect measurements that are traditionally used in biomolecular structure determination by NMR. In a de novo structure determination, however, lack of knowledge about the degree and orientation of molecular alignment complicates the analysis of dipolar coupling data. We present a probabilistic framework for analyzing residual dipolar couplings and demonstrate that it is possible to estimate the atomic coordinates, the complete molecular alignment tensor, and the error of the couplings simultaneously. As a by-product, we also obtain estimates of the uncertainty in the coordinates and the alignment tensor. We show that our approach encompasses existing methods for determining the alignment tensor as special cases, including least squares estimation, histogram fitting, and elimination of an explicit alignment tensor in the restraint energy.

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Contour-propagation Algorithms for Semi-automated Reconstruction of Neural Processes

Macke, J., Maack, N., Gupta, R., Denk, W., Schölkopf, B., Borst, A.

Journal of Neuroscience Methods, 167(2):349-357, January 2008 (article)

Abstract
A new technique, ”Serial Block Face Scanning Electron Microscopy” (SBFSEM), allows for automatic sectioning and imaging of biological tissue with a scanning electron microscope. Image stacks generated with this technology have a resolution sufficient to distinguish different cellular compartments, including synaptic structures, which should make it possible to obtain detailed anatomical knowledge of complete neuronal circuits. Such an image stack contains several thousands of images and is recorded with a minimal voxel size of 10-20nm in the x and y- and 30nm in z-direction. Consequently, a tissue block of 1mm3 (the approximate volume of the Calliphora vicina brain) will produce several hundred terabytes of data. Therefore, highly automated 3D reconstruction algorithms are needed. As a first step in this direction we have developed semiautomated segmentation algorithms for a precise contour tracing of cell membranes. These algorithms were embedded into an easy-to-operate user interface, which allows direct 3D observation of the extracted objects during the segmentation of image stacks. Compared to purely manual tracing, processing time is greatly accelerated.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
A Quantum-Statistical-Mechanical Extension of Gaussian Mixture Model

Tanaka, K., Tsuda, K.

Journal of Physics: Conference Series, 95(012023):1-9, January 2008 (article)

Abstract
We propose an extension of Gaussian mixture models in the statistical-mechanical point of view. The conventional Gaussian mixture models are formulated to divide all points in given data to some kinds of classes. We introduce some quantum states constructed by superposing conventional classes in linear combinations. Our extension can provide a new algorithm in classifications of data by means of linear response formulas in the statistical mechanics.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Learning to control in operational space

Peters, J., Schaal, S.

International Journal of Robotics Research, 27, pages: 197-212, 2008, clmc (article)

Abstract
One of the most general frameworks for phrasing control problems for complex, redundant robots is operational space control. However, while this framework is of essential importance for robotics and well-understood from an analytical point of view, it can be prohibitively hard to achieve accurate control in face of modeling errors, which are inevitable in com- plex robots, e.g., humanoid robots. In this paper, we suggest a learning approach for opertional space control as a direct inverse model learning problem. A first important insight for this paper is that a physically cor- rect solution to the inverse problem with redundant degrees-of-freedom does exist when learning of the inverse map is performed in a suitable piecewise linear way. The second crucial component for our work is based on the insight that many operational space controllers can be understood in terms of a constrained optimal control problem. The cost function as- sociated with this optimal control problem allows us to formulate a learn- ing algorithm that automatically synthesizes a globally consistent desired resolution of redundancy while learning the operational space controller. From the machine learning point of view, this learning problem corre- sponds to a reinforcement learning problem that maximizes an immediate reward. We employ an expectation-maximization policy search algorithm in order to solve this problem. Evaluations on a three degrees of freedom robot arm are used to illustrate the suggested approach. The applica- tion to a physically realistic simulator of the anthropomorphic SARCOS Master arm demonstrates feasibility for complex high degree-of-freedom robots. We also show that the proposed method works in the setting of learning resolved motion rate control on real, physical Mitsubishi PA-10 medical robotics arm.

link (url) DOI [BibTex]

link (url) DOI [BibTex]

2006


no image
Some observations on the pedestal effect or dipper function

Henning, B., Wichmann, F.

Journal of Vision, 6(13):50, 2006 Fall Vision Meeting of the Optical Society of America, December 2006 (poster)

Abstract
The pedestal effect is the large improvement in the detectabilty of a sinusoidal “signal” grating observed when the signal is added to a masking or “pedestal” grating of the same spatial frequency, orientation, and phase. We measured the pedestal effect in both broadband and notched noise - noise from which a 1.5-octave band centred on the signal frequency had been removed. Although the pedestal effect persists in broadband noise, it almost disappears in the notched noise. Furthermore, the pedestal effect is substantial when either high- or low-pass masking noise is used. We conclude that the pedestal effect in the absence of notched noise results principally from the use of information derived from channels with peak sensitivities at spatial frequencies different from that of the signal and pedestal. The spatial-frequency components of the notched noise above and below the spatial frequency of the signal and pedestal prevent the use of information about changes in contrast carried in channels tuned to spatial frequencies that are very much different from that of the signal and pedestal. Thus the pedestal or dipper effect measured without notched noise is not a characteristic of individual spatial-frequency tuned channels.

Web DOI [BibTex]

2006

Web DOI [BibTex]


no image
A Kernel Method for the Two-Sample-Problem

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

20th Annual Conference on Neural Information Processing Systems (NIPS), December 2006 (talk)

Abstract
We propose two statistical tests to determine if two samples are from different distributions. Our test statistic is in both cases the distance between the means of the two samples mapped into a reproducing kernel Hilbert space (RKHS). The first test is based on a large deviation bound for the test statistic, while the second is based on the asymptotic distribution of this statistic. We show that the test statistic can be computed in $O(m^2)$ time. We apply our approach to a variety of problems, including attribute matching for databases using the Hungarian marriage method, where our test performs strongly. We also demonstrate excellent performance when comparing distributions over graphs, for which no alternative tests currently exist.

PDF [BibTex]

PDF [BibTex]


no image
Ab-initio gene finding using machine learning

Schweikert, G., Zeller, G., Zien, A., Ong, C., de Bona, F., Sonnenburg, S., Phillips, P., Rätsch, G.

NIPS Workshop on New Problems and Methods in Computational Biology, December 2006 (talk)

Web [BibTex]

Web [BibTex]


no image
Graph boosting for molecular QSAR analysis

Saigo, H., Kadowaki, T., Kudo, T., Tsuda, K.

NIPS Workshop on New Problems and Methods in Computational Biology, December 2006 (talk)

Abstract
We propose a new boosting method that systematically combines graph mining and mathematical programming-based machine learning. Informative and interpretable subgraph features are greedily found by a series of graph mining calls. Due to our mathematical programming formulation, subgraph features and pre-calculated real-valued features are seemlessly integrated. We tested our algorithm on a quantitative structure-activity relationship (QSAR) problem, which is basically a regression problem when given a set of chemical compounds. In benchmark experiments, the prediction accuracy of our method favorably compared with the best results reported on each dataset.

Web [BibTex]

Web [BibTex]


no image
Inferring Causal Directions by Evaluating the Complexity of Conditional Distributions

Sun, X., Janzing, D., Schölkopf, B.

NIPS Workshop on Causality and Feature Selection, December 2006 (talk)

Abstract
We propose a new approach to infer the causal structure that has generated the observed statistical dependences among n random variables. The idea is that the factorization of the joint measure of cause and effect into P(cause)P(effect|cause) leads typically to simpler conditionals than non-causal factorizations. To evaluate the complexity of the conditionals we have tried two methods. First, we have compared them to those which maximize the conditional entropy subject to the observed first and second moments since we consider the latter as the simplest conditionals. Second, we have fitted the data with conditional probability measures being exponents of functions in an RKHS space and defined the complexity by a Hilbert-space semi-norm. Such a complexity measure has several properties that are useful for our purpose. We describe some encouraging results with both methods applied to real-world data. Moreover, we have combined constraint-based approaches to causal discovery (i.e., methods using only information on conditional statistical dependences) with our method in order to distinguish between causal hypotheses which are equivalent with respect to the imposed independences. Furthermore, we compare the performance to Bayesian approaches to causal inference.

Web [BibTex]


no image
Structure validation of the Josephin domain of ataxin-3: Conclusive evidence for an open conformation

Nicastro, G., Habeck, M., Masino, L., Svergun, DI., Pastore, A.

Journal of Biomolecular NMR, 36(4):267-277, December 2006 (article)

Abstract
The availability of new and fast tools in structure determination has led to a more than exponential growth of the number of structures solved per year. It is therefore increasingly essential to assess the accuracy of the new structures by reliable approaches able to assist validation. Here, we discuss a specific example in which the use of different complementary techniques, which include Bayesian methods and small angle scattering, resulted essential for validating the two currently available structures of the Josephin domain of ataxin-3, a protein involved in the ubiquitin/proteasome pathway and responsible for neurodegenerative spinocerebellar ataxia of type 3. Taken together, our results demonstrate that only one of the two structures is compatible with the experimental information. Based on the high precision of our refined structure, we show that Josephin contains an open cleft which could be directly implicated in the interaction with polyubiquitin chains and other partners.

Web DOI [BibTex]

Web DOI [BibTex]


no image
A Unifying View of Wiener and Volterra Theory and Polynomial Kernel Regression

Franz, M., Schölkopf, B.

Neural Computation, 18(12):3097-3118, December 2006 (article)

Abstract
Volterra and Wiener series are perhaps the best understood nonlinear system representations in signal processing. Although both approaches have enjoyed a certain popularity in the past, their application has been limited to rather low-dimensional and weakly nonlinear systems due to the exponential growth of the number of terms that have to be estimated. We show that Volterra and Wiener series can be represented implicitly as elements of a reproducing kernel Hilbert space by utilizing polynomial kernels. The estimation complexity of the implicit representation is linear in the input dimensionality and independent of the degree of nonlinearity. Experiments show performance advantages in terms of convergence, interpretability, and system sizes that can be handled.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Learning Optimal EEG Features Across Time, Frequency and Space

Farquhar, J., Hill, J., Schölkopf, B.

NIPS Workshop on Current Trends in Brain-Computer Interfacing, December 2006 (talk)

PDF Web [BibTex]

PDF Web [BibTex]


no image
Semi-Supervised Learning

Zien, A.

Advanced Methods in Sequence Analysis Lectures, November 2006 (talk)

Web [BibTex]

Web [BibTex]


no image
Statistical Analysis of Slow Crack Growth Experiments

Pfingsten, T., Glien, K.

Journal of the European Ceramic Society, 26(15):3061-3065, November 2006 (article)

Abstract
A common approach for the determination of Slow Crack Growth (SCG) parameters are the static and dynamic loading method. Since materials with small Weibull module show a large variability in strength, a correct statistical analysis of the data is indispensable. In this work we propose the use of the Maximum Likelihood method and a Baysian analysis, which, in contrast to the standard procedures, take into account that failure strengths are Weibull distributed. The analysis provides estimates for the SCG parameters, the Weibull module, and the corresponding confidence intervals and overcomes the necessity of manual differentiation between inert and fatigue strength data. We compare the methods to a Least Squares approach, which can be considered the standard procedure. The results for dynamic loading data from the glass sealing of MEMS devices show that the assumptions inherent to the standard approach lead to significantly different estimates.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Optimizing Spatial Filters for BCI: Margin- and Evidence-Maximization Approaches

Farquhar, J., Hill, N., Schölkopf, B.

Challenging Brain-Computer Interfaces: MAIA Workshop 2006, pages: 1, November 2006 (poster)

Abstract
We present easy-to-use alternatives to the often-used two-stage Common Spatial Pattern + classifier approach for spatial filtering and classification of Event-Related Desychnronization signals in BCI. We report two algorithms that aim to optimize the spatial filters according to a criterion more directly related to the ability of the algorithms to generalize to unseen data. Both are based upon the idea of treating the spatial filter coefficients as hyperparameters of a kernel or covariance function. We then optimize these hyper-parameters directly along side the normal classifier parameters with respect to our chosen learning objective function. The two objectives considered are margin maximization as used in Support-Vector Machines and the evidence maximization framework used in Gaussian Processes. Our experiments assessed generalization error as a function of the number of training points used, on 9 BCI competition data sets and 5 offline motor imagery data sets measured in Tubingen. Both our approaches sho w consistent improvements relative to the commonly used CSP+linear classifier combination. Strikingly, the improvement is most significant in the higher noise cases, when either few trails are used for training, or with the most poorly performing subjects. This a reversal of the usual "rich get richer" effect in the development of CSP extensions, which tend to perform best when the signal is strong enough to accurately find their additional parameters. This makes our approach particularly suitable for clinical application where high levels of noise are to be expected.

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
A Machine Learning Approach for Determining the PET Attenuation Map from Magnetic Resonance Images

Hofmann, M., Steinke, F., Judenhofer, M., Claussen, C., Schölkopf, B., Pichler, B.

IEEE Medical Imaging Conference, November 2006 (talk)

Abstract
A promising new combination in multimodality imaging is MR-PET, where the high soft tissue contrast of Magnetic Resonance Imaging (MRI) and the functional information of Positron Emission Tomography (PET) are combined. Although many technical problems have recently been solved, it is still an open problem to determine the attenuation map from the available MR scan, as the MR intensities are not directly related to the attenuation values. One standard approach is an atlas registration where the atlas MR image is aligned with the patient MR thus also yielding an attenuation image for the patient. We also propose another approach, which to our knowledge has not been tried before: Using Support Vector Machines we predict the attenuation value directly from the local image information. We train this well-established machine learning algorithm using small image patches. Although both approaches sometimes yielded acceptable results, they also showed their specific shortcomings: The registration often fails with large deformations whereas the prediction approach is problematic when the local image structure is not characteristic enough. However, the failures often do not coincide and integration of both information sources is promising. We therefore developed a combination method extending Support Vector Machines to use not only local image structure but also atlas registered coordinates. We demonstrate the strength of this combination approach on a number of examples.

[BibTex]

[BibTex]


no image
Mining frequent stem patterns from unaligned RNA sequences

Hamada, M., Tsuda, K., Kudo, T., Kin, T., Asai, K.

Bioinformatics, 22(20):2480-2487, October 2006 (article)

Abstract
Motivation: In detection of non-coding RNAs, it is often necessary to identify the secondary structure motifs from a set of putative RNA sequences. Most of the existing algorithms aim to provide the best motif or few good motifs, but biologists often need to inspect all the possible motifs thoroughly. Results: Our method RNAmine employs a graph theoretic representation of RNA sequences, and detects all the possible motifs exhaustively using a graph mining algorithm. The motif detection problem boils down to finding frequently appearing patterns in a set of directed and labeled graphs. In the tasks of common secondary structure prediction and local motif detection from long sequences, our method performed favorably both in accuracy and in efficiency with the state-of-the-art methods such as CMFinder.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Large-Scale Gene Expression Profiling Reveals Major Pathogenetic Pathways of Cartilage Degeneration in Osteoarthritis

Aigner, T., Fundel, K., Saas, J., Gebhard, P., Haag, J., Weiss, T., Zien, A., Obermayr, F., Zimmer, R., Bartnik, E.

Arthritis and Rheumatism, 54(11):3533-3544, October 2006 (article)

Abstract
Objective. Despite many research efforts in recent decades, the major pathogenetic mechanisms of osteo- arthritis (OA), including gene alterations occurring during OA cartilage degeneration, are poorly under- stood, and there is no disease-modifying treatment approach. The present study was therefore initiated in order to identify differentially expressed disease-related genes and potential therapeutic targets. Methods. This investigation consisted of a large gene expression profiling study performed based on 78 normal and disease samples, using a custom-made complementar y DNA array covering >4,000 genes. Results. Many differentially expressed genes were identified, including the expected up-regulation of ana- bolic and catabolic matrix genes. In particular, the down-regulation of important oxidative defense genes, i.e., the genes for superoxide dismutases 2 and 3 and glutathione peroxidase 3, was prominent. This indicates that continuous oxidative stress to the cells and the matrix is one major underlying pathogenetic mecha- nism in OA. Also, genes that are involved in the phenot ypic stabilit y of cells, a feature that is greatly reduced in OA cartilage, appeared to be suppressed. Conclusion. Our findings provide a reference data set on gene alterations in OA cartilage and, importantly, indicate major mechanisms underlying central cell bio- logic alterations that occur during the OA disease process. These results identify molecular targets that can be further investigated in the search for therapeutic interventions.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Semi-Supervised Support Vector Machines and Application to Spam Filtering

Zien, A.

ECML Discovery Challenge Workshop, September 2006 (talk)

Abstract
After introducing the semi-supervised support vector machine (aka TSVM for "transductive SVM"), a few popular training strategies are briefly presented. Then the assumptions underlying semi-supervised learning are reviewed. Finally, two modern TSVM optimization techniques are applied to the spam filtering data sets of the workshop; it is shown that they can achieve excellent results, if the problem of the data being non-iid can be handled properly.

PDF Web [BibTex]


no image
Learning Eye Movements

Kienzle, W., Wichmann, F., Schölkopf, B., Franz, M.

Sensory Coding And The Natural Environment, 2006, pages: 1, September 2006 (poster)

Abstract
The human visual system samples images through saccadic eye movements which rapidly change the point of fixation. Although the selection of eye movement targets depends on numerous top-down mechanisms, a number of recent studies have shown that low-level image features such as local contrast or edges play an important role. These studies typically used predefined image features which were afterwards experimentally verified. Here, we follow a complementary approach: instead of testing a set of candidate image features, we infer these hypotheses from the data, using methods from statistical learning. To this end, we train a non-linear classifier on fixated vs. randomly selected image patches without making any physiological assumptions. The resulting classifier can be essentially characterized by a nonlinear combination of two center-surround receptive fields. We find that the prediction performance of this simple model on our eye movement data is indistinguishable from the physiologically motivated model of Itti & Koch (2000) which is far more complex. In particular, we obtain a comparable performance without using any multi-scale representations, long-range interactions or oriented image features.

Web [BibTex]

Web [BibTex]


no image
Implicit Surface Modelling with a Globally Regularised Basis of Compact Support

Walder, C., Schölkopf, B., Chapelle, O.

Computer Graphics Forum, 25(3):635-644, September 2006 (article)

Abstract
We consider the problem of constructing a globally smooth analytic function that represents a surface implicitly by way of its zero set, given sample points with surface normal vectors. The contributions of the paper include a novel means of regularising multi-scale compactly supported basis functions that leads to the desirable interpolation properties previously only associated with fully supported bases. We also provide a regularisation framework for simpler and more direct treatment of surface normals, along with a corresponding generalisation of the representer theorem lying at the core of kernel-based machine learning methods. We demonstrate the techniques on 3D problems of up to 14 million data points, as well as 4D time series data and four-dimensional interpolation between three-dimensional shapes.

PDF GZIP DOI [BibTex]


no image
Inferential Structure Determination: Probabilistic determination and validation of NMR structures

Habeck, M.

Gordon Research Conference on Computational Aspects of Biomolecular NMR, September 2006 (talk)

Web [BibTex]

Web [BibTex]


no image
An Online Support Vector Machine for Abnormal Events Detection

Davy, M., Desobry, F., Gretton, A., Doncarli, C.

Signal Processing, 86(8):2009-2025, August 2006 (article)

Abstract
The ability to detect online abnormal events in signals is essential in many real-world Signal Processing applications. Previous algorithms require an explicit signal statistical model, and interpret abnormal events as statistical model abrupt changes. Corresponding implementation relies on maximum likelihood or on Bayes estimation theory with generally excellent performance. However, there are numerous cases where a robust and tractable model cannot be obtained, and model-free approaches need to be considered. In this paper, we investigate a machine learning, descriptor-based approach that does not require an explicit descriptors statistical model, based on Support Vector novelty detection. A sequential optimization algorithm is introduced. Theoretical considerations as well as simulations on real signals demonstrate its practical efficiency.

PDF PostScript PDF DOI [BibTex]

PDF PostScript PDF DOI [BibTex]


no image
Machine Learning Algorithms for Polymorphism Detection

Schweikert, G., Zeller, G., Clark, R., Ossowski, S., Warthmann, N., Shinn, P., Frazer, K., Ecker, J., Huson, D., Weigel, D., Schölkopf, B., Rätsch, G.

2nd ISCB Student Council Symposium, August 2006 (talk)

Abstract
Analyzing resequencing array data using machine learning, we obtain a genome-wide inventory of polymorphisms in 20 wild strains of Arabidopsis thaliana, including 750,000 single nucleotide poly- morphisms (SNPs) and thousands of highly polymorphic regions and deletions. We thus provide an unprecedented resource for the study of natural variation in plants.

Web [BibTex]

Web [BibTex]


no image
Integrating Structured Biological data by Kernel Maximum Mean Discrepancy

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

Bioinformatics, 22(4: ISMB 2006 Conference Proceedings):e49-e57, August 2006 (article)

Abstract
Motivation: Many problems in data integration in bioinformatics can be posed as one common question: Are two sets of observations generated by the same distribution? We propose a kernel-based statistical test for this problem, based on the fact that two distributions are different if and only if there exists at least one function having different expectation on the two distributions. Consequently we use the maximum discrepancy between function means as the basis of a test statistic. The Maximum Mean Discrepancy (MMD) can take advantage of the kernel trick, which allows us to apply it not only to vectors, but strings, sequences, graphs, and other common structured data types arising in molecular biology. Results: We study the practical feasibility of an MMD-based test on three central data integration tasks: Testing cross-platform comparability of microarray data, cancer diagnosis, and data-content based schema matching for two different protein function classification schemas. In all of these experiments, including high-dimensional ones, MMD is very accurate in finding samples that were generated from the same distribution, and outperforms its best competitors. Conclusions: We have defined a novel statistical test of whether two samples are from the same distribution, compatible with both multivariate and structured data, that is fast, easy to implement, and works well, as confirmed by our experiments.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Large Scale Transductive SVMs

Collobert, R., Sinz, F., Weston, J., Bottou, L.

Journal of Machine Learning Research, 7, pages: 1687-1712, August 2006 (article)

Abstract
We show how the Concave-Convex Procedure can be applied to the optimization of Transductive SVMs, which traditionally requires solving a combinatorial search problem. This provides for the first time a highly scalable algorithm in the nonlinear case. Detailed experiments verify the utility of our approach.

PostScript PDF PDF [BibTex]

PostScript PDF PDF [BibTex]


no image
Building Support Vector Machines with Reduced Classifier Complexity

Keerthi, S., Chapelle, O., DeCoste, D.

Journal of Machine Learning Research, 7, pages: 1493-1515, July 2006 (article)

Abstract
Support vector machines (SVMs), though accurate, are not preferred in applications requiring great classification speed, due to the number of support vectors being large. To overcome this problem we devise a primal method with the following properties: (1) it decouples the idea of basis functions from the concept of support vectors; (2) it greedily finds a set of kernel basis functions of a specified maximum size ($dmax$) to approximate the SVM primal cost function well; (3) it is efficient and roughly scales as $O(ndmax^2)$ where $n$ is the number of training examples; and, (4) the number of basis functions it requires to achieve an accuracy close to the SVM accuracy is usually far less than the number of SVM support vectors.

PDF [BibTex]

PDF [BibTex]


no image
Inferential structure determination: Overview and new developments

Habeck, M.

Sixth CCPN Annual Conference: Efficient and Rapid Structure Determination by NMR, July 2006 (talk)

Web [BibTex]

Web [BibTex]


no image
ARTS: Accurate Recognition of Transcription Starts in Human

Sonnenburg, S., Zien, A., Rätsch, G.

Bioinformatics, 22(14):e472-e480, July 2006 (article)

Abstract
Motivation: One of the most important features of genomic DNA are the protein-coding genes. While it is of great value to identify those genes and the encoded proteins, it is also crucial to understand how their transcription is regulated. To this end one has to identify the corresponding promoters and the contained transcription factor binding sites. TSS finders can be used to locate potential promoters. They may also be used in combination with other signal and content detectors to resolve entire gene structures. Results: We have developed a novel kernel based method - called ARTS - that accurately recognizes transcription start sites in human. The application of otherwise too computationally expensive Support Vector Machines was made possible due to the use of efficient training and evaluation techniques using suffix tries. In a carefully designed experimental study, we compare our TSS finder to state-of-the-art methods from the literature: McPromoter, Eponine and FirstEF. For given false positive rates within a reasonable range, we consistently achieve considerably higher true positive rates. For instance, ARTS finds about 24% true positives at a false positive rate of 1/1000, where the other methods find less than half (10.5%). Availability: Datasets, model selection results, whole genome predictions, and additional experimental results are available at http://www.fml.tuebingen.mpg.de/raetsch/projects/arts

Web DOI [BibTex]

Web DOI [BibTex]


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]

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
Classification of natural scenes: Critical features revisited

Drewes, J., Wichmann, F., Gegenfurtner, K.

Journal of Vision, 6(6):561, 6th Annual Meeting of the Vision Sciences Society (VSS), June 2006 (poster)

Abstract
Human observers are capable of detecting animals within novel natural scenes with remarkable speed and accuracy. Despite the seeming complexity of such decisions it has been hypothesized that a simple global image feature, the relative abundance of high spatial frequencies at certain orientations, could underly such fast image classification (A. Torralba & A. Oliva, Network: Comput. Neural Syst., 2003). We successfully used linear discriminant analysis to classify a set of 11.000 images into “animal” and “non-animal” images based on their individual amplitude spectra only (Drewes, Wichmann, Gegenfurtner VSS 2005). We proceeded to sort the images based on the performance of our classifier, retaining only the best and worst classified 400 images (“best animals”, “best distractors” and “worst animals”, “worst distractors”). We used a Go/No-go paradigm to evaluate human performance on this subset of our images. Both reaction time and proportion of correctly classified images showed a significant effect of classification difficulty. Images more easily classified by our algorithm were also classified faster and better by humans, as predicted by the Torralba & Oliva hypothesis. We then equated the amplitude spectra of the 400 images, which, by design, reduced algorithmic performance to chance whereas human performance was only slightly reduced (cf. Wichmann, Rosas, Gegenfurtner, VSS 2005). Most importantly, the same images as before were still classified better and faster, suggesting that even in the original condition features other than specifics of the amplitude spectrum made particular images easy to classify, clearly at odds with the Torralba & Oliva hypothesis.

Web DOI [BibTex]

Web DOI [BibTex]


no image
MCMC inference in (Conditionally) Conjugate Dirichlet Process Gaussian Mixture Models

Rasmussen, C., Görür, D.

ICML Workshop on Learning with Nonparametric Bayesian Methods, June 2006 (talk)

Abstract
We compare the predictive accuracy of the Dirichlet Process Gaussian mixture models using conjugate and conditionally conjugate priors and show that better density models result from using the wider class of priors. We explore several MCMC schemes exploiting conditional conjugacy and show their computational merits on several multidimensional density estimation problems.

Web [BibTex]

Web [BibTex]


no image
Sampling for non-conjugate infinite latent feature models

Görür, D., Rasmussen, C.

(Editors: Bernardo, J. M.), 8th Valencia International Meeting on Bayesian Statistics (ISBA), June 2006 (talk)

Abstract
Latent variable models are powerful tools to model the underlying structure in data. Infinite latent variable models can be defined using Bayesian nonparametrics. Dirichlet process (DP) models constitute an example of infinite latent class models in which each object is assumed to belong to one of the, mutually exclusive, infinitely many classes. Recently, the Indian buffet process (IBP) has been defined as an extension of the DP. IBP is a distribution over sparse binary matrices with infinitely many columns which can be used as a distribution for non-exclusive features. Inference using Markov chain Monte Carlo (MCMC) in conjugate IBP models has been previously described, however requiring conjugacy restricts the use of IBP. We describe an MCMC algorithm for non-conjugate IBP models. Modelling the choice behaviour is an important topic in psychology, economics and related fields. Elimination by Aspects (EBA) is a choice model that assumes each alternative has latent features with associated weights that lead to the observed choice outcomes. We formulate a non-parametric version of EBA by using IBP as the prior over the latent binary features. We infer the features of objects that lead to the choice data by using our sampling scheme for inference.

PDF [BibTex]

PDF [BibTex]