Header logo is ei


2007


no image
On the Representer Theorem and Equivalent Degrees of Freedom of SVR

Dinuzzo, F., Neve, M., De Nicolao, G., Gianazza, U.

Journal of Machine Learning Research, 8, pages: 2467-2495, October 2007 (article)

Abstract
Support Vector Regression (SVR) for discrete data is considered. An alternative formulation of the representer theorem is derived. This result is based on the newly introduced notion of pseudoresidual and the use of subdifferential calculus. The representer theorem is exploited to analyze the sensitivity properties of ε-insensitive SVR and introduce the notion of approximate degrees of freedom. The degrees of freedom are shown to play a key role in the evaluation of the optimism, that is the difference between the expected in-sample error and the expected empirical risk. In this way, it is possible to define a Cp-like statistic that can be used for tuning the parameters of SVR. The proposed tuning procedure is tested on a simulated benchmark problem and on a real world problem (Boston Housing data set).

Web [BibTex]

2007

Web [BibTex]


no image
A Hilbert Space Embedding for Distributions

Smola, A., Gretton, A., Song, L., Schölkopf, B.

Proceedings of the 10th International Conference on Discovery Science (DS 2007), 10, pages: 40-41, October 2007 (poster)

Abstract
While kernel methods are the basis of many popular techniques in supervised learning, they are less commonly used in testing, estimation, and analysis of probability distributions, where information theoretic approaches rule the roost. However it becomes difficult to estimate mutual information or entropy if the data are high dimensional.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Unsupervised Classification for non-invasive Brain-Computer-Interfaces

Eren, S., Grosse-Wentrup, M., Buss, M.

In Automed 2007, pages: 65-66, VDI Verlag, Düsseldorf, Germany, Automed Workshop, October 2007 (inproceedings)

Abstract
Non-invasive Brain-Computer-Interfaces (BCIs) are devices that infer the intention of human subjects from signals generated by the central nervous system and recorded outside the skull, e.g., by electroencephalography (EEG). They can be used to enable basic communication for patients who are not able to communicate by normal means, e.g., due to neuro-degenerative diseases such as amyotrophic lateral sclerosis (ALS) (see [Vaughan2003] for a review). One challenge in research on BCIs is minimizing the training time prior to usage of the BCI. Since EEG patterns vary across subjects, it is usually necessary to record a number of trials in which the intention of the user is known to train a classifier. This classifier is subsequently used to infer the intention of the BCI-user. In this paper, we present the application of an unsupervised classification method to a binary noninvasive BCI based on motor imagery. The result is a BCI that does not require any training, since the mapping from EEG pattern changes to the intention of the user is learned online by the BCI without any feedback. We present experimental results from six healthy subjects, three of which display classification errors below 15%. We conclude that unsupervised BCIs are a viable option, but not yet as reliable as supervised BCIs. The rest of this paper is organized as follows. In the Methods section, we first introduce the experimental paradigm. This is followed by a description of the methods used for spatial filtering, feature extraction, and unsupervised classification. We then present the experimental results, and conclude the paper with a brief discussion.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Some observations on the masking effects of Mach bands

Curnow, T., Cowie, DA., Henning, GB., Hill, NJ.

Journal of the Optical Society of America A, 24(10):3233-3241, October 2007 (article)

Abstract
There are 8 cycle / deg ripples or oscillations in performance as a function of location near Mach bands in experiments measuring Mach bands’ masking effects on random polarity signal bars. The oscillations with increments are 180 degrees out of phase with those for decrements. The oscillations, much larger than the measurement error, appear to relate to the weighting function of the spatial-frequency-tuned channel detecting the broad- band signals. The ripples disappear with step maskers and become much smaller at durations below 25 ms, implying either that the site of masking has changed or that the weighting function and hence spatial-frequency tuning is slow to develop.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
A Hilbert Space Embedding for Distributions

Smola, A., Gretton, A., Song, L., Schölkopf, B.

In Algorithmic Learning Theory, Lecture Notes in Computer Science 4754 , pages: 13-31, (Editors: M Hutter and RA Servedio and E Takimoto), Springer, Berlin, Germany, 18th International Conference on Algorithmic Learning Theory (ALT), October 2007 (inproceedings)

Abstract
We describe a technique for comparing distributions without the need for density estimation as an intermediate step. Our approach relies on mapping the distributions into a reproducing kernel Hilbert space. Applications of this technique can be found in two-sample tests, which are used for determining whether two sets of observations arise from the same distribution, covariate shift correction, local learning, measures of independence, and density estimation.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


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]

PDF [BibTex]


no image
Cluster Identification in Nearest-Neighbor Graphs

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

In ALT 2007, pages: 196-210, (Editors: Hutter, M. , R. A. Servedio, E. Takimoto), Springer, Berlin, Germany, 18th International Conference on Algorithmic Learning Theory, October 2007 (inproceedings)

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 PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Support Vector Machine Learning for Interdependent and Structured Output Spaces

Altun, Y., Hofmann, T., Tsochantaridis, I.

In Predicting Structured Data, pages: 85-104, Advances in neural information processing systems, (Editors: Bakir, G. H. , T. Hofmann, B. Schölkopf, A. J. Smola, B. Taskar, S. V. N. Vishwanathan), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Web [BibTex]

Web [BibTex]


no image
Inducing Metric Violations in Human Similarity Judgements

Laub, J., Macke, J., Müller, K., Wichmann, F.

In Advances in Neural Information Processing Systems 19, pages: 777-784, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
Attempting to model human categorization and similarity judgements is both a very interesting but also an exceedingly difficult challenge. Some of the difficulty arises because of conflicting evidence whether human categorization and similarity judgements should or should not be modelled as to operate on a mental representation that is essentially metric. Intuitively, this has a strong appeal as it would allow (dis)similarity to be represented geometrically as distance in some internal space. Here we show how a single stimulus, carefully constructed in a psychophysical experiment, introduces l2 violations in what used to be an internal similarity space that could be adequately modelled as Euclidean. We term this one influential data point a conflictual judgement. We present an algorithm of how to analyse such data and how to identify the crucial point. Thus there may not be a strict dichotomy between either a metric or a non-metric internal space but rather degrees to which potentially large subsets of stimuli are represented metrically with a small subset causing a global violation of metricity.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Cross-Validation Optimization for Large Scale Hierarchical Classification Kernel Methods

Seeger, M.

In Advances in Neural Information Processing Systems 19, pages: 1233-1240, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We propose a highly efficient framework for kernel multi-class models with a large and structured set of classes. Kernel parameters are learned automatically by maximizing the cross-validation log likelihood, and predictive probabilities are estimated. We demonstrate our approach on large scale text classification tasks with hierarchical class structure, achieving state-of-the-art results in an order of magnitude less time than previous work.

PDF Web [BibTex]

PDF Web [BibTex]


no image
A Local Learning Approach for Clustering

Wu, M., Schölkopf, B.

In Advances in Neural Information Processing Systems 19, pages: 1529-1536, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We present a local learning approach for clustering. The basic idea is that a good clustering result should have the property that the cluster label of each data point can be well predicted based on its neighboring data and their cluster labels, using current supervised learning methods. An optimization problem is formulated such that its solution has the above property. Relaxation and eigen-decomposition are applied to solve this optimization problem. We also briefly investigate the parameter selection issue and provide a simple parameter selection method for the proposed algorithm. Experimental results are provided to validate the effectiveness of the proposed approach.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Brisk Kernel ICA

Jegelka, S., Gretton, A.

In Large Scale Kernel Machines, pages: 225-250, Neural Information Processing, (Editors: Bottou, L. , O. Chapelle, D. DeCoste, J. Weston), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Abstract
Recent approaches to independent component analysis have used kernel independence measures to obtain very good performance in ICA, particularly in areas where classical methods experience difficulty (for instance, sources with near-zero kurtosis). In this chapter, we compare two efficient extensions of these methods for large-scale problems: random subsampling of entries in the Gram matrices used in defining the independence measures, and incomplete Cholesky decomposition of these matrices. We derive closed-form, efficiently computable approximations for the gradients of these measures, and compare their performance on ICA using both artificial and music data. We show that kernel ICA can scale up to much larger problems than yet attempted, and that incomplete Cholesky decomposition performs better than random sampling.

PDF Web [BibTex]

PDF Web [BibTex]


no image
MR-Based PET Attenuation Correction: Method and Validation

Hofmann, M., Steinke, F., Scheel, V., Brady, M., Schölkopf, B., Pichler, B.

Joint Molecular Imaging Conference, September 2007 (talk)

Abstract
PET/MR combines the high soft tissue contrast of Magnetic Resonance Imaging (MRI) and the functional information of Positron Emission Tomography (PET). For quantitative PET information, correction of tissue photon attenuation is mandatory. Usually in conventional PET, the attenuation map is obtained from a transmission scan, which uses a rotating source, or from the CT scan in case of combined PET/CT. In the case of a PET/MR scanner, there is insufficient space for the rotating source and ideally one would want to calculate the attenuation map from the MR image instead. Since MR images provide information about proton density of the different tissue types, it is not trivial to use this data for PET attenuation correction. We present a method for predicting the PET attenuation map from a given the MR image, using a combination of atlas-registration and recognition of local patterns. Using "leave one out cross validation" we show on a database of 16 MR-CT image pairs that our method reliably allows estimating the CT image from the MR image. Subsequently, as in PET/CT, the PET attenuation map can be predicted from the CT image. On an additional dataset of MR/CT/PET triplets we quantitatively validate that our approach allows PET quantification with an error that is smaller than what would be clinically significant. We demonstrate our approach on T1-weighted human brain scans. However, the presented methods are more general and current research focuses on applying the established methods to human whole body PET/MRI applications.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Adaptive Spatial Filters with predefined Region of Interest for EEG based Brain-Computer-Interfaces

Grosse-Wentrup, M., Gramann, K., Buss, M.

In Advances in Neural Information Processing Systems 19, pages: 537-544, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
The performance of EEG-based Brain-Computer-Interfaces (BCIs) critically depends on the extraction of features from the EEG carrying information relevant for the classification of different mental states. For BCIs employing imaginary movements of different limbs, the method of Common Spatial Patterns (CSP) has been shown to achieve excellent classification results. The CSP-algorithm however suffers from a lack of robustness, requiring training data without artifacts for good performance. To overcome this lack of robustness, we propose an adaptive spatial filter that replaces the training data in the CSP approach by a-priori information. More specifically, we design an adaptive spatial filter that maximizes the ratio of the variance of the electric field originating in a predefined region of interest (ROI) and the overall variance of the measured EEG. Since it is known that the component of the EEG used for discriminating imaginary movements originates in the motor cortex, we design two adaptive spatial filters with the ROIs centered in the hand areas of the left and right motor cortex. We then use these to classify EEG data recorded during imaginary movements of the right and left hand of three subjects, and show that the adaptive spatial filters outperform the CSP-algorithm, enabling classification rates of up to 94.7 % without artifact rejection.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Predicting Structured Data

Bakir, G., Hofmann, T., Schölkopf, B., Smola, A., Taskar, B., Vishwanathan, S.

pages: 360, Advances in neural information processing systems, MIT Press, Cambridge, MA, USA, September 2007 (book)

Abstract
Machine learning develops intelligent computer systems that are able to generalize from previously seen examples. A new domain of machine learning, in which the prediction must satisfy the additional constraints found in structured data, poses one of machine learning’s greatest challenges: learning functional dependencies between arbitrary input and output domains. This volume presents and analyzes the state of the art in machine learning algorithms and theory in this novel field. The contributors discuss applications as diverse as machine translation, document markup, computational biology, and information extraction, among others, providing a timely overview of an exciting field.

Web [BibTex]

Web [BibTex]


no image
Mining complex genotypic features for predicting HIV-1 drug resistance

Saigo, H., Uno, T., Tsuda, K.

Bioinformatics, 23(18):2455-2462, September 2007 (article)

Abstract
Human immunodeficiency virus type 1 (HIV-1) evolves in human body, and its exposure to a drug often causes mutations that enhance the resistance against the drug. To design an effective pharmacotherapy for an individual patient, it is important to accurately predict the drug resistance based on genotype data. Notably, the resistance is not just the simple sum of the effects of all mutations. Structural biological studies suggest that the association of mutations is crucial: Even if mutations A or B alone do not affect the resistance, a significant change might happen when the two mutations occur together. Linear regression methods cannot take the associations into account, while decision tree methods can reveal only limited associations. Kernel methods and neural networks implicitly use all possible associations for prediction, but cannot select salient associations explicitly. Our method, itemset boosting, performs linear regression in the complete space of power sets of mutations. It implements a forward feature selection procedure where, in each iteration, one mutation combination is found by an efficient branch-and-bound search. This method uses all possible combinations, and salient associations are explicitly shown. In experiments, our method worked particularly well for predicting the resistance of nucleotide reverse transcriptase inhibitors (NRTIs). Furthermore, it successfully recovered many mutation associations known in biological literature.

Web DOI [BibTex]

Web DOI [BibTex]


no image
Branch and Bound for Semi-Supervised Support Vector Machines

Chapelle, O., Sindhwani, V., Keerthi, S.

In Advances in Neural Information Processing Systems 19, pages: 217-224, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
Semi-supervised SVMs (S3VMs) attempt to learn low-density separators by maximizing the margin over labeled and unlabeled examples. The associated optimization problem is non-convex. To examine the full potential of S3VMs modulo local minima problems in current implementations, we apply branch and bound techniques for obtaining exact, globally optimal solutions. Empirical evidence suggests that the globally optimal solution can return excellent generalization performance in situations where other implementations fail completely. While our current implementation is only applicable to small datasets, we discuss variants that can potentially lead to practically useful algorithms.

PDF Web [BibTex]

PDF Web [BibTex]


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

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

In Advances in Neural Information Processing Systems 19, pages: 513-520, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

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. 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 Web [BibTex]

PDF Web [BibTex]


no image
An Efficient Method for Gradient-Based Adaptation of Hyperparameters in SVM Models

Keerthi, S., Sindhwani, V., Chapelle, O.

In Advances in Neural Information Processing Systems 19, pages: 673-680, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We consider the task of tuning hyperparameters in SVM models based on minimizing a smooth performance validation function, e.g., smoothed k-fold cross-validation error, using non-linear optimization techniques. The key computation in this approach is that of the gradient of the validation function with respect to hyperparameters. We show that for large-scale problems involving a wide choice of kernel-based models and validation functions, this computation can be very efficiently done; often within just a fraction of the training time. Empirical results show that a near-optimal set of hyperparameters can be identified by our approach with very few training rounds and gradient computations.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Learning Dense 3D Correspondence

Steinke, F., Schölkopf, B., Blanz, V.

In Advances in Neural Information Processing Systems 19, pages: 1313-1320, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
Establishing correspondence between distinct objects is an important and nontrivial task: correctness of the correspondence hinges on properties which are difficult to capture in an a priori criterion. While previous work has used a priori criteria which in some cases led to very good results, the present paper explores whether it is possible to learn a combination of features that, for a given training set of aligned human heads, characterizes the notion of correct correspondence. By optimizing this criterion, we are then able to compute correspondence and morphs for novel heads.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Optimal Dominant Motion Estimation using Adaptive Search of Transformation Space

Ulges, A., Lampert, CH., Keysers, D., Breuel, TM.

In DAGM 2007, pages: 204-215, (Editors: Hamprecht, F. A., C. Schnörr, B. Jähne), Springer, Berlin, Germany, 29th Annual Symposium of the German Association for Pattern Recognition, September 2007 (inproceedings)

Abstract
The extraction of a parametric global motion from a motion field is a task with several applications in video processing. We present two probabilistic formulations of the problem and carry out optimization using the RAST algorithm, a geometric matching method novel to motion estimation in video. RAST uses an exhaustive and adaptive search of transformation space and thus gives -- in contrast to local sampling optimization techniques used in the past -- a globally optimal solution. Among other applications, our framework can thus be used as a source of ground truth for benchmarking motion estimation algorithms. Our main contributions are: first, the novel combination of a state-of- the-art MAP criterion for dominant motion estimation with a search procedure that guarantees global optimality. Second, experimental re- sults that illustrate the superior performance of our approach on synthetic flow fields as well as real-world video streams. Third, a significant speedup of the search achieved by extending the mod el with an additional smoothness prior.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Training a Support Vector Machine in the Primal

Chapelle, O.

In Large Scale Kernel Machines, pages: 29-50, Neural Information Processing, (Editors: Bottou, L. , O. Chapelle, D. DeCoste, J. Weston), MIT Press, Cambridge, MA, USA, September 2007, This is a slightly updated version of the Neural Computation paper (inbook)

Abstract
Most literature on Support Vector Machines (SVMs) concentrate on the dual optimization problem. In this paper, we would like to point out that the primal problem can also be solved efficiently, both for linear and non-linear SVMs, and that there is no reason to ignore this possibility. On the contrary, from the primal point of view new families of algorithms for large scale SVM training can be investigated.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Approximation Methods for Gaussian Process Regression

Quiñonero-Candela, J., Rasmussen, CE., Williams, CKI.

In Large-Scale Kernel Machines, pages: 203-223, Neural Information Processing, (Editors: Bottou, L. , O. Chapelle, D. DeCoste, J. Weston), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Abstract
A wealth of computationally efficient approximation methods for Gaussian process regression have been recently proposed. We give a unifying overview of sparse approximations, following Quiñonero-Candela and Rasmussen (2005), and a brief review of approximate matrix-vector multiplication methods.

PDF Web [BibTex]

PDF Web [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
Studying the effects of noise correlations on population coding using a sampling method

Ecker, A., Berens, P., Bethge, M., Logothetis, N., Tolias, A.

Neural Coding, Computation and Dynamics (NCCD 07), 1, pages: 21, September 2007 (poster)

PDF [BibTex]

PDF [BibTex]


no image
Large-Scale Kernel Machines

Bottou, L., Chapelle, O., DeCoste, D., Weston, J.

pages: 416, Neural Information Processing Series, MIT Press, Cambridge, MA, USA, September 2007 (book)

Abstract
Pervasive and networked computers have dramatically reduced the cost of collecting and distributing large datasets. In this context, machine learning algorithms that scale poorly could simply become irrelevant. We need learning algorithms that scale linearly with the volume of the data while maintaining enough statistical efficiency to outperform algorithms that simply process a random subset of the data. This volume offers researchers and engineers practical solutions for learning from large scale datasets, with detailed descriptions of algorithms and experiments carried out on realistically large datasets. At the same time it offers researchers information that can address the relative lack of theoretical grounding for many useful algorithms. After a detailed description of state-of-the-art support vector machine technology, an introduction of the essential concepts discussed in the volume, and a comparison of primal and dual optimization techniques, the book progresses from well-understood techniques to more novel and controversial approaches. Many contributors have made their code and data available online for further experimentation. Topics covered include fast implementations of known algorithms, approximations that are amenable to theoretical guarantees, and algorithms that perform well in practice but are difficult to analyze theoretically.

Web [BibTex]

Web [BibTex]


no image
Solving Deep Memory POMDPs with Recurrent Policy Gradients

Wierstra, D., Förster, A., Peters, J., Schmidhuber, J.

In ICANN‘07, pages: 697-706, Springer, Berlin, Germany, International Conference on Artificial Neural Networks, September 2007 (inproceedings)

Abstract
This paper presents Recurrent Policy Gradients, a modelfree reinforcement learning (RL) method creating limited-memory stochastic policies for partially observable Markov decision problems (POMDPs) that require long-term memories of past observations. The approach involves approximating a policy gradient for a Recurrent Neural Network (RNN) by backpropagating return-weighted characteristic eligibilities through time. Using a “Long Short-Term Memory” architecture, we are able to outperform other RL methods on two important benchmark tasks. Furthermore, we show promising results on a complex car driving simulation task.

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Density Estimation of Structured Outputs in Reproducing Kernel Hilbert Spaces

Altun, Y., Smola, A.

In Predicting Structured Data, pages: 283-300, Advances in neural information processing systems, (Editors: BakIr, G. H., T. Hofmann, B. Schölkopf, A. J. Smola, B. Taskar, S. V.N. Vishwanathan), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Abstract
In this paper we study the problem of estimating conditional probability distributions for structured output prediction tasks in Reproducing Kernel Hilbert Spaces. More specically, we prove decomposition results for undirected graphical models, give constructions for kernels, and show connections to Gaussian Process classi- cation. Finally we present ecient means of solving the optimization problem and apply this to label sequence learning. Experiments on named entity recognition and pitch accent prediction tasks demonstrate the competitiveness of our approach.

Web [BibTex]

Web [BibTex]


no image
Advances in Neural Information Processing Systems 19: Proceedings of the 2006 Conference

Schölkopf, B., Platt, J., Hofmann, T.

Proceedings of the Twentieth Annual Conference on Neural Information Processing Systems (NIPS 2006), pages: 1690, MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (proceedings)

Abstract
The annual Neural Information Processing Systems (NIPS) conference is the flagship meeting on neural computation and machine learning. It draws a diverse group of attendees--physicists, neuroscientists, mathematicians, statisticians, and computer scientists--interested in theoretical and applied aspects of modeling, simulating, and building neural-like or intelligent systems. The presentations are interdisciplinary, with contributions in algorithms, learning theory, cognitive science, neuroscience, brain imaging, vision, speech and signal processing, reinforcement learning, and applications. Only twenty-five percent of the papers submitted are accepted for presentation at NIPS, so the quality is exceptionally high. This volume contains the papers presented at the December 2006 meeting, held in Vancouver.

Web [BibTex]

Web [BibTex]


no image
Output Grouping using Dirichlet Mixtures of Linear Gaussian State-Space Models

Chiappa, S., Barber, D.

In ISPA 2007, pages: 446-451, IEEE Computer Society, Los Alamitos, CA, USA, 5th International Symposium on Image and Signal Processing and Analysis, September 2007 (inproceedings)

Abstract
We consider a model to cluster the components of a vector time-series. The task is to assign each component of the vector time-series to a single cluster, basing this assignment on the simultaneous dynamical similarity of the component to other components in the cluster. This is in contrast to the more familiar task of clustering a set of time-series based on global measures of their similarity. The model is based on a Dirichlet Mixture of Linear Gaussian State-Space models (LGSSMs), in which each LGSSM is treated with a prior to encourage the simplest explanation. The resulting model is approximated using a ‘collapsed’ variational Bayes implementation.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Manifold Denoising

Hein, M., Maier, M.

In Advances in Neural Information Processing Systems 19, pages: 561-568, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We consider the problem of denoising a noisily sampled submanifold $M$ in $R^d$, where the submanifold $M$ is a priori unknown and we are only given a noisy point sample. The presented denoising algorithm is based on a graph-based diffusion process of the point sample. We analyze this diffusion process using recent results about the convergence of graph Laplacians. In the experiments we show that our method is capable of dealing with non-trivial high-dimensional noise. Moreover using the denoising algorithm as pre-processing method we can improve the results of a semi-supervised learning algorithm.

PDF Web [BibTex]

PDF Web [BibTex]


no image
How to Find Interesting Locations in Video: A Spatiotemporal Interest Point Detector Learned from Human Eye movements

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

In Pattern Recognition, pages: 405-414, (Editors: FA Hamprecht and C Schnörr and B Jähne), Springer, Berlin, Germany, 29th Annual Symposium of the German Association for Pattern Recognition (DAGM), September 2007 (inproceedings)

Abstract
Interest point detection in still images is a well-studied topic in computer vision. In the spatiotemporal domain, however, it is still unclear which features indicate useful interest points. In this paper we approach the problem by emph{learning} a detector from examples: we record eye movements of human subjects watching video sequences and train a neural network to predict which locations are likely to become eye movement targets. We show that our detector outperforms current spatiotemporal interest point architectures on a standard classification dataset.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Bayesian Inference for Sparse Generalized Linear Models

Seeger, M., Gerwinn, S., Bethge, M.

In ECML 2007, pages: 298-309, Lecture Notes in Computer Science ; 4701, (Editors: Kok, J. N., J. Koronacki, R. Lopez de Mantaras, S. Matwin, D. Mladenic, A. Skowron), Springer, Berlin, Germany, 18th European Conference on Machine Learning, September 2007 (inproceedings)

Abstract
We present a framework for efficient, accurate approximate Bayesian inference in generalized linear models (GLMs), based on the expectation propagation (EP) technique. The parameters can be endowed with a factorizing prior distribution, encoding properties such as sparsity or non-negativity. The central role of posterior log-concavity in Bayesian GLMs is emphasized and related to stability issues in EP. In particular, we use our technique to infer the parameters of a point process model for neuronal spiking data from multiple electrodes, demonstrating significantly superior predictive performance when a sparsity assumption is enforced via a Laplace prior distribution.

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Implicit Surfaces with Globally Regularised and Compactly Supported Basis Functions

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

In Advances in Neural Information Processing Systems 19, pages: 273-280, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We consider the problem of constructing a function whose zero set is to represent a surface, given sample points with surface normal vectors. The contributions include a novel means of regularising multi-scale compactly supported basis functions that leads to the desirable properties previously only associated with fully supported bases, and show equivalence to a Gaussian process with modified covariance function. We also provide a regularisation framework for simpler and more direct treatment of surface normals, along with a corresponding generalisation of the representer theorem. We demonstrate the techniques on 3D problems of up to 14 million data points, as well as 4D time series data.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Trading Convexity for Scalability

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

In Large Scale Kernel Machines, pages: 275-300, Neural Information Processing, (Editors: Bottou, L. , O. Chapelle, D. DeCoste, J. Weston), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Abstract
Convex learning algorithms, such as Support Vector Machines (SVMs), are often seen as highly desirable because they offer strong practical properties and are amenable to theoretical analysis. However, in this work we show how nonconvexity can provide scalability advantages over convexity. We show how concave-convex programming can be applied to produce (i) faster SVMs where training errors are no longer support vectors, and (ii) much faster Transductive SVMs.

PDF Web [BibTex]

PDF Web [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
Bayesian methods for NMR structure determination

Habeck, M.

29th Annual Discussion Meeting: Magnetic Resonance in Biophysical Chemistry, September 2007 (talk)

Web [BibTex]

Web [BibTex]


no image
Comparison of Adaptive Spatial Filters with Heuristic and Optimized Region of Interest for EEG Based Brain-Computer-Interfaces

Liefhold, C., Grosse-Wentrup, M., Gramann, K., Buss, M.

In Pattern Recognition, pages: 274-283, (Editors: Hamprecht, F. A., C. Schnörr, B. Jähne), Springer, Berlin, Germany, 29th Annual Symposium of the German Association for Pattern Recognition, September 2007 (inproceedings)

Abstract
Research on EEG based brain-computer-interfaces (BCIs) aims at steering devices by thought. Even for simple applications, BCIs require an extremely effective data processing to work properly because of the low signal-to-noise-ratio (SNR) of EEG signals. Spatial filtering is one successful preprocessing method, which extracts EEG components carrying the most relevant information. Unlike spatial filtering with Common Spatial Patterns (CSP), Adaptive Spatial Filtering (ASF) can be adapted to freely selectable regions of interest (ROI) and with this, artifacts can be actively suppressed. In this context, we compare the performance of ASF with ROIs selected using anatomical a-priori information and ASF with numerically optimized ROIs. Therefore, we introduce a method for data driven spatial filter adaptation and apply the achieved filters for classification of EEG data recorded during imaginary movements of the left and right hand of four subjects. The results show, that in the case of artifact-free datasets, ASFs with numerically optimized ROIs achieve classification rates of up to 97.7 % while ASFs with ROIs defined by anatomical heuristic stay at 93.7 % for the same data. Otherwise, with noisy datasets, the former brake down (66.7 %) while the latter meet 95.7 %.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Machine Learning and Applications in Biology

Shin, H.

In BioKorea 2007, pages: 337-366, BioKorea, September 2007 (inproceedings)

Abstract
The emergence of the fields of computational biology and bioinformatics has alleviated the burden of solving many biological problems, saving the time and cost required for experiments and also providing predictions that guide new experiments. Within computational biology, machine learning algorithms have played a central role in dealing with the flood of biological data. The goal of this tutorial is to raise awareness and comprehension of machine learning so that biologists can properly match the task at hand to the corresponding analytical approach. We start by categorizing biological problem settings and introduce the general machine learning schemes that fit best to each or these categories. We then explore representative models in further detail, from traditional statistical models to recent kernel models, presenting several up-to-date research projects in bioinfomatics to exemplify how biological questions can benefit from a machine learning approach. Finally, we discuss how cooperation between biologis ts and machine learners might be made smoother.

[BibTex]

[BibTex]


no image
Real-Time Fetal Heart Monitoring in Biomagnetic Measurements Using Adaptive Real-Time ICA

Waldert, S., Bensch, M., Bogdan, M., Rosenstiel, W., Schölkopf, B., Lowery, C., Eswaran, H., Preissl, H.

IEEE Transactions on Biomedical Engineering, 54(10):1867-1874, September 2007 (article)

Abstract
Electrophysiological signals of the developing fetal brain and heart can be investigated by fetal magnetoencephalography (fMEG). During such investigations, the fetal heart activity and that of the mother should be monitored continuously to provide an important indication of current well-being. Due to physical constraints of an fMEG system, it is not possible to use clinically established heart monitors for this purpose. Considering this constraint, we developed a real-time heart monitoring system for biomagnetic measurements and showed its reliability and applicability in research and for clinical examinations. The developed system consists of real-time access to fMEG data, an algorithm based on Independent Component Analysis (ICA), and a graphical user interface (GUI). The algorithm extracts the current fetal and maternal heart signal from a noisy and artifact-contaminated data stream in real-time and is able to adapt automatically to continuously varying environmental parameters. This algorithm has been na med Adaptive Real-time ICA (ARICA) and is applicable to real-time artifact removal as well as to related blind signal separation problems.

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
A Nonparametric Approach to Bottom-Up Visual Saliency

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

In Advances in Neural Information Processing Systems 19, pages: 689-696, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
This paper addresses the bottom-up influence of local image information on human eye movements. Most existing computational models use a set of biologically plausible linear filters, e.g., Gabor or Difference-of-Gaussians filters as a front-end, the outputs of which are nonlinearly combined into a real number that indicates visual saliency. Unfortunately, this requires many design parameters such as the number, type, and size of the front-end filters, as well as the choice of nonlinearities, weighting and normalization schemes etc., for which biological plausibility cannot always be justified. As a result, these parameters have to be chosen in a more or less ad hoc way. Here, we propose to emph{learn} a visual saliency model directly from human eye movement data. The model is rather simplistic and essentially parameter-free, and therefore contrasts recent developments in the field that usually aim at higher prediction rates at the cost of additional parameters and increasing model complexity. Experimental results show that - despite the lack of any biological prior knowledge - our model performs comparably to existing approaches, and in fact learns image features that resemble findings from several previous studies. In particular, its maximally excitatory stimuli have center-surround structure, similar to receptive fields in the early human visual system.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Information Bottleneck for Non Co-Occurrence Data

Seldin, Y., Slonim, N., Tishby, N.

In Advances in Neural Information Processing Systems 19, pages: 1241-1248, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We present a general model-independent approach to the analysis of data in cases when these data do not appear in the form of co-occurrence of two variables X, Y, but rather as a sample of values of an unknown (stochastic) function Z(X,Y). For example, in gene expression data, the expression level Z is a function of gene X and condition Y; or in movie ratings data the rating Z is a function of viewer X and movie Y . The approach represents a consistent extension of the Information Bottleneck method that has previously relied on the availability of co-occurrence statistics. By altering the relevance variable we eliminate the need in the sample of joint distribution of all input variables. This new formulation also enables simple MDL-like model complexity control and prediction of missing values of Z. The approach is analyzed and shown to be on a par with the best known clustering algorithms for a wide range of domains. For the prediction of missing values (collaborative filtering) it improves the currently best known results.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Learning with Hypergraphs: Clustering, Classification, and Embedding

Zhou, D., Huang, J., Schölkopf, B.

In Advances in Neural Information Processing Systems 19, pages: 1601-1608, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We usually endow the investigated objects with pairwise relationships, which can be illustrated as graphs. In many real-world problems, however, relationships among the objects of our interest are more complex than pairwise. Naively squeezing the complex relationships into pairwise ones will inevitably lead to loss of information which can be expected valuable for our learning tasks however. Therefore we consider using hypergraphs instead to completely represent complex relationships among the objects of our interest, and thus the problem of learning with hypergraphs arises. Our main contribution in this paper is to generalize the powerful methodology of spectral clustering which originally operates on undirected graphs to hypergraphs, and further develop algorithms for hypergraph embedding and transductive classi¯cation on the basis of the spectral hypergraph clustering approach. Our experiments on a number of benchmarks showed the advantages of hypergraphs over usual graphs.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Classifying Event-Related Desynchronization in EEG, ECoG and MEG signals

Hill, N., Lal, T., Tangermann, M., Hinterberger, T., Widman, G., Elger, C., Schölkopf, B., Birbaumer, N.

In Toward Brain-Computer Interfacing, pages: 235-260, Neural Information Processing, (Editors: G Dornhege and J del R Millán and T Hinterberger and DJ McFarland and K-R Müller), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

PDF Web [BibTex]

PDF Web [BibTex]


no image
Fast Computation of Graph Kernels

Vishwanathan, SVN., Borgwardt, KM., Schraudolph, N.

In Advances in Neural Information Processing Systems 19, pages: 1449-1456, (Editors: Schölkopf, B. , J. Platt, T. Hofmann), MIT Press, Cambridge, MA, USA, Twentieth Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
Using extensions of linear algebra concepts to Reproducing Kernel Hilbert Spaces (RKHS), we define a unifying framework for random walk kernels on graphs. Reduction to a Sylvester equation allows us to compute many of these kernels in O(n3) worst-case time. This includes kernels whose previous worst-case time complexity was O(n6), such as the geometric kernels of G¨artner et al. [1] and the marginal graph kernels of Kashima et al. [2]. Our algebra in RKHS allow us to exploit sparsity in directed and undirected graphs more effectively than previous methods, yielding sub-cubic computational complexity when combined with conjugate gradient solvers or fixed-point iterations. Experiments on graphs from bioinformatics and other application domains show that our algorithms are often more than 1000 times faster than existing approaches.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Joint Kernel Maps

Weston, J., Bakir, G., Bousquet, O., Mann, T., Noble, W., Schölkopf, B.

In Predicting Structured Data, pages: 67-84, Advances in neural information processing systems, (Editors: GH Bakir and T Hofmann and B Schölkopf and AJ Smola and B Taskar and SVN Vishwanathan), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

Web [BibTex]

Web [BibTex]


no image
Correcting Sample Selection Bias by Unlabeled Data

Huang, J., Smola, A., Gretton, A., Borgwardt, K., Schölkopf, B.

In Advances in Neural Information Processing Systems 19, pages: 601-608, (Editors: B Schölkopf and J Platt and T Hofmann), MIT Press, Cambridge, MA, USA, 20th Annual Conference on Neural Information Processing Systems (NIPS), September 2007 (inproceedings)

Abstract
We consider the scenario where training and test data are drawn from different distributions, commonly referred to as sample selection bias. Most algorithms for this setting try to first recover sampling distributions and then make appropriate corrections based on the distribution estimate. We present a nonparametric method which directly produces resampling weights without distribution estimation. Our method works by matching distributions between training and testing sets in feature space. Experimental results demonstrate that our method works well in practice.

PDF Web [BibTex]

PDF Web [BibTex]


no image
Brain-Computer Interfaces for Communication in Paralysis: A Clinical Experimental Approach

Hinterberger, T., Nijboer, F., Kübler, A., Matuz, T., Furdea, A., Mochty, U., Jordan, M., Lal, T., Hill, J., Mellinger, J., Bensch, M., Tangermann, M., Widman, G., Elger, C., Rosenstiel, W., Schölkopf, B., Birbaumer, N.

In Toward Brain-Computer Interfacing, pages: 43-64, Neural Information Processing, (Editors: G. Dornhege and J del R Millán and T Hinterberger and DJ McFarland and K-R Müller), MIT Press, Cambridge, MA, USA, September 2007 (inbook)

PDF Web [BibTex]

PDF Web [BibTex]


no image
Near-Maximum Entropy Models for Binary Neural Representations of Natural Images

Berens, P., Bethge, M.

Neural Coding, Computation and Dynamics (NCCD 07), 1, pages: 19, September 2007 (poster)

Abstract
Maximum entropy analysis of binary variables provides an elegant way for studying the role of pairwise correlations in neural populations. Unfortunately, these approaches suffer from their poor scalability to high dimensions. In sensory coding, however, high-dimensional data is ubiquitous. Here, we introduce a new approach using a near-maximum entropy model, that makes this type of analysis feasible for very high-dimensional data---the model parameters can be derived in closed form and sampling is easy. We demonstrate its usefulness by studying a simple neural representation model of natural images. For the first time, we are able to directly compare predictions from a pairwise maximum entropy model not only in small groups of neurons, but also in larger populations of more than thousand units. Our results indicate that in such larger networks interactions exist that are not predicted by pairwise correlations, despite the fact that pairwise correlations explain the lower-dimensional marginal statistics extrem ely well up to the limit of dimensionality where estimation of the full joint distribution is feasible.

PDF [BibTex]

PDF [BibTex]