Header logo is


2019


no image
Semi-supervised learning, causality, and the conditional cluster assumption

von Kügelgen, J., Mey, A., Loog, M., Schölkopf, B.

NeurIPS 2019 Workshop “Do the right thing”: machine learning and causal inference for improved decision making, December 2019 (poster) Accepted

ei

link (url) [BibTex]

2019


link (url) [BibTex]


no image
Optimal experimental design via Bayesian optimization: active causal structure learning for Gaussian process networks

von Kügelgen, J., Rubenstein, P., Schölkopf, B., Weller, A.

NeurIPS 2019 Workshop “Do the right thing”: machine learning and causal inference for improved decision making, December 2019 (poster) Accepted

ei

link (url) [BibTex]

link (url) [BibTex]


no image
Convolutional neural networks: A magic bullet for gravitational-wave detection?

Gebhard, T., Kilbertus, N., Harry, I., Schölkopf, B.

Physical Review D, 100(6):063015, American Physical Society, September 2019 (article)

ei

link (url) DOI [BibTex]

link (url) DOI [BibTex]


no image
Data scarcity, robustness and extreme multi-label classification

Babbar, R., Schölkopf, B.

Machine Learning, 108(8):1329-1351, September 2019, Special Issue of the ECML PKDD 2019 Journal Track (article)

ei

DOI [BibTex]

DOI [BibTex]


no image
A 32-channel multi-coil setup optimized for human brain shimming at 9.4T

Aghaeifar, A., Zhou, J., Heule, R., Tabibian, B., Schölkopf, B., Jia, F., Zaitsev, M., Scheffler, K.

Magnetic Resonance in Medicine, 2019, (Early View) (article)

ei

DOI [BibTex]

DOI [BibTex]


no image
Enhancing Human Learning via Spaced Repetition Optimization

Tabibian, B., Upadhyay, U., De, A., Zarezade, A., Schölkopf, B., Gomez Rodriguez, M.

Proceedings of the National Academy of Sciences, 2019, PNAS published ahead of print January 22, 2019 (article)

ei

DOI Project Page Project Page [BibTex]

DOI Project Page Project Page [BibTex]


Thumb xl screenshot 2019 03 25 at 14.29.22
Learning to Control Highly Accelerated Ballistic Movements on Muscular Robots

Büchler, D., Calandra, R., Peters, J.

2019 (article) Submitted

Abstract
High-speed and high-acceleration movements are inherently hard to control. Applying learning to the control of such motions on anthropomorphic robot arms can improve the accuracy of the control but might damage the system. The inherent exploration of learning approaches can lead to instabilities and the robot reaching joint limits at high speeds. Having hardware that enables safe exploration of high-speed and high-acceleration movements is therefore desirable. To address this issue, we propose to use robots actuated by Pneumatic Artificial Muscles (PAMs). In this paper, we present a four degrees of freedom (DoFs) robot arm that reaches high joint angle accelerations of up to 28000 °/s^2 while avoiding dangerous joint limits thanks to the antagonistic actuation and limits on the air pressure ranges. With this robot arm, we are able to tune control parameters using Bayesian optimization directly on the hardware without additional safety considerations. The achieved tracking performance on a fast trajectory exceeds previous results on comparable PAM-driven robots. We also show that our system can be controlled well on slow trajectories with PID controllers due to careful construction considerations such as minimal bending of cables, lightweight kinematics and minimal contact between PAMs and PAMs with the links. Finally, we propose a novel technique to control the the co-contraction of antagonistic muscle pairs. Experimental results illustrate that choosing the optimal co-contraction level is vital to reach better tracking performance. Through the use of PAM-driven robots and learning, we do a small step towards the future development of robots capable of more human-like motions.

ei

Arxiv Video [BibTex]


no image
Perception of temporal dependencies in autoregressive motion

Meding, K., Schölkopf, B., Wichmann, F. A.

European Conference on Visual Perception (ECVP), 2019 (poster)

ei

[BibTex]

[BibTex]


no image
Inferring causation from time series with perspectives in Earth system sciences

Runge, J., Bathiany, S., Bollt, E., Camps-Valls, G., Coumou, D., Deyle, E., Glymour, C., Kretschmer, M., Mahecha, M., van Nes, E., Peters, J., Quax, R., Reichstein, M., Scheffer, M. S. B., Spirtes, P., Sugihara, G., Sun, J., Zhang, K., Zscheischler, J.

Nature Communications, 2019 (article) In revision

ei

[BibTex]

[BibTex]


no image
Phenomenal Causality and Sensory Realism

Bruijns, S. A., Meding, K., Schölkopf, B., Wichmann, F. A.

European Conference on Visual Perception (ECVP), 2019 (poster)

ei

[BibTex]

[BibTex]

2010


no image
Similarities in resting state and feature-driven activity: Non-parametric evaluation of human fMRI

Shelton, J., Blaschko, M., Gretton, A., Müller, J., Fischer, E., Bartels, A.

NIPS Workshop on Learning and Planning from Batch Time Series Data, December 2010 (poster)

ei

PDF Web [BibTex]

2010


PDF Web [BibTex]


no image
Causal relationships between frequency bands of extracellular signals in visual cortex revealed by an information theoretic analysis

Besserve, M., Schölkopf, B., Logothetis, N., Panzeri, S.

Journal of Computational Neuroscience, 29(3):547-566, December 2010 (article)

ei

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Tackling Box-Constrained Optimization via a New Projected Quasi-Newton Approach

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

SIAM Journal on Scientific Computing, 32(6):3548-3563 , December 2010 (article)

Abstract
Numerous scientific applications across a variety of fields depend on box-constrained convex optimization. Box-constrained problems therefore continue to attract research interest. We address box-constrained (strictly convex) problems by deriving two new quasi-Newton algorithms. Our algorithms are positioned between the projected-gradient [J. B. Rosen, J. SIAM, 8 (1960), pp. 181–217] and projected-Newton [D. P. Bertsekas, SIAM J. Control Optim., 20 (1982), pp. 221–246] methods. We also prove their convergence under a simple Armijo step-size rule. We provide experimental results for two particular box-constrained problems: nonnegative least squares (NNLS), and nonnegative Kullback–Leibler (NNKL) minimization. For both NNLS and NNKL our algorithms perform competitively as compared to well-established methods on medium-sized problems; for larger problems our approach frequently outperforms the competition.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Algorithmen zum Automatischen Erlernen von Motorfähigkeiten

Peters, J., Kober, J., Schaal, S.

at - Automatisierungstechnik, 58(12):688-694, December 2010 (article)

Abstract
Robot learning methods which allow autonomous robots to adapt to novel situations have been a long standing vision of robotics, artificial intelligence, and cognitive sciences. However, to date, learning techniques have yet to fulfill this promise as only few methods manage to scale into the high-dimensional domains of manipulator robotics, or even the new upcoming trend of humanoid robotics. If possible, scaling was usually only achieved in precisely pre-structured domains. In this paper, we investigate the ingredients for a general approach policy learning with the goal of an application to motor skill refinement in order to get one step closer towards human-like performance. For doing so, we study two major components for such an approach, i. e., firstly, we study policy learning algorithms which can be applied in the general setting of motor skill learning, and, secondly, we study a theoretically well-founded general approach to representing the required control structures for task representation and execution.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
PAC-Bayesian Analysis of Co-clustering and Beyond

Seldin, Y., Tishby, N.

Journal of Machine Learning Research, 11, pages: 3595-3646, December 2010 (article)

ei

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
Augmentation of fMRI Data Analysis using Resting State Activity and Semi-supervised Canonical Correlation Analysis

Shelton, JA., Blaschko, MB., Bartels, A.

NIPS Women in Machine Learning Workshop (WiML), December 2010 (poster)

Abstract
Resting state activity is brain activation that arises in the absence of any task, and is usually measured in awake subjects during prolonged fMRI scanning sessions where the only instruction given is to close the eyes and do nothing. It has been recognized in recent years that resting state activity is implicated in a wide variety of brain function. While certain networks of brain areas have different levels of activation at rest and during a task, there is nevertheless significant similarity between activations in the two cases. This suggests that recordings of resting state activity can be used as a source of unlabeled data to augment kernel canonical correlation analysis (KCCA) in a semisupervised setting. We evaluate this setting empirically yielding three main results: (i) KCCA tends to be improved by the use of Laplacian regularization even when no additional unlabeled data are available, (ii) resting state data seem to have a similar marginal distribution to that recorded during the execution of a visual processing task implying largely similar types of activation, and (iii) this source of information can be broadly exploited to improve the robustness of empirical inference in fMRI studies, an inherently data poor domain.

ei

PDF Web [BibTex]

PDF Web [BibTex]


no image
Gaussian Processes for Machine Learning (GPML) Toolbox

Rasmussen, C., Nickisch, H.

Journal of Machine Learning Research, 11, pages: 3011-3015, November 2010 (article)

Abstract
The GPML toolbox provides a wide range of functionality for Gaussian process (GP) inference and prediction. GPs are specified by mean and covariance functions; we offer a library of simple mean and covariance functions and mechanisms to compose more complex ones. Several likelihood functions are supported including Gaussian and heavy-tailed for regression as well as others suitable for classification. Finally, a range of inference methods is provided, including exact and variational inference, Expectation Propagation, and Laplace's method dealing with non-Gaussian likelihoods and FITC for dealing with large regression tasks.

ei

Web [BibTex]

Web [BibTex]


no image
Cryo-EM structure and rRNA model of a translating eukaryotic 80S ribosome at 5.5-Å resolution

Armache, J-P., Jarasch, A., Anger, AM., Villa, E., Becker, T., Bhushan, S., Jossinet, F., Habeck, M., Dindar, G., Franckenberg, S., Marquez, V., Mielke, T., Thomm, M., Berninghausen, O., Beatrix, B., Söding, J., Westhof, E., Wilson, DN., Beckmann, R.

Proceedings of the National Academy of Sciences of the United States of America, 107(46):19748-19753, November 2010 (article)

Abstract
Protein biosynthesis, the translation of the genetic code into polypeptides, occurs on ribonucleoprotein particles called ribosomes. Although X-ray structures of bacterial ribosomes are available, high-resolution structures of eukaryotic 80S ribosomes are lacking. Using cryoelectron microscopy and single-particle reconstruction, we have determined the structure of a translating plant (Triticum aestivum) 80S ribosome at 5.5-Å resolution. This map, together with a 6.1-Å map of a Saccharomyces cerevisiae 80S ribosome, has enabled us to model ∼98% of the rRNA. Accurate assignment of the rRNA expansion segments (ES) and variable regions has revealed unique ES–ES and r-protein–ES interactions, providing insight into the structure and evolution of the eukaryotic ribosome.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Policy gradient methods

Peters, J.

Scholarpedia, 5(11):3698, November 2010 (article)

Abstract
Policy gradient methods are a type of reinforcement learning techniques that rely upon optimizing parametrized policies with respect to the expected return (long-term cumulative reward) by gradient descent. They do not suffer from many of the problems that have been marring traditional reinforcement learning approaches such as the lack of guarantees of a value function, the intractability problem resulting from uncertain state information and the complexity arising from continuous states & actions.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
High frequency phase-spike synchronization of extracellular signals modulates causal interactions in monkey primary visual cortex

Besserve, M., Murayama, Y., Schölkopf, B., Logothetis, N., Panzeri, S.

40(616.2), 40th Annual Meeting of the Society for Neuroscience (Neuroscience), November 2010 (poster)

Abstract
Functional correlates of Rhythms in the gamma band (30-100Hz) are observed in the mammalian brain with a large variety of functional correlates. Nevertheless, their functional role is still debated. One way to disentangle this issue is to go beyond usual correlation analysis and apply causality measures that quantify the directed interactions between the gamma rhythms and other aspects of neural activity. These measures can be further compared with other aspects of neurophysicological signals to find markers of neural interactions. In a recent study, we analyzed extracellular recordings in the primary visual cortex of 4 anesthetized macaques during the presentation of movie stimuli using a causality measure named Transfer Entropy. We found causal interactions between high frequency gamma rhythms (60-100Hz) recorded in different electrodes, involving in particular their phase, and between the gamma phase and spiking activity quantified by the instantaneous envelope of the MUA band (1-3kHz). Here, we further investigate in the same dataset the meaning of these phase-MUA and phase-phase causal interactions by studying the distribution of phases at multiple recording sites at lags around the occurrence of spiking events. First, we found a sharpening of the gamma phase distribution in one electrode when spikes are occurring in other recording site. This phenomena appeared as a form of phase-spike synchronization and was quantified by an information theoretic measure. We found this measure correlates significantly with phase-MUA causal interactions. Additionally, we quantified in a similar way the interplay between spiking and the phase difference between two recording sites (reflecting the well-know concept of phase synchronization). We found that, depending on the couple of recording site, spiking can correlate either with a phase synchronization or with a desynchronization with respect to the baseline. This effect correlates very well with the phase-phase causality measure. These results provide evidence for high frequency phase-spike synchronization to reflect communication between distant neural populations in V1. Conversely, both phase synchronization or desynchronization may favor neural communication between recording sites. This new result, which contrasts with current hypothesis on the role of phase synchronization, could be interpreted as the presence of inhibitory interactions that are suppressed by desynchronization. Finally, our findings give new insights into the role of gamma rhythms in regulating local computation in the visual cortex.

ei

Web [BibTex]

Web [BibTex]


no image
Localization of eukaryote-specific ribosomal proteins in a 5.5-Å cryo-EM map of the 80S eukaryotic ribosome

Armache, J-P., Jarasch, A., Anger, AM., Villa, E., Becker, T., Bhushan, S., Jossinet, F., Habeck, M., Dindar, G., Franckenberg, S., Marquez, V., Mielke, T., Thomm, M., Berninghausen, O., Beatrix, B., Söding, J., Westhof, E., Wilson, DN., Beckmann, R.

Proceedings of the National Academy of Sciences of the United States of America, 107(46):19754-19759, November 2010 (article)

Abstract
Protein synthesis in all living organisms occurs on ribonucleoprotein particles, called ribosomes. Despite the universality of this process, eukaryotic ribosomes are significantly larger in size than their bacterial counterparts due in part to the presence of 80 r proteins rather than 54 in bacteria. Using cryoelectron microscopy reconstructions of a translating plant (Triticum aestivum) 80S ribosome at 5.5-Å resolution, together with a 6.1-Å map of a translating Saccharomyces cerevisiae 80S ribosome, we have localized and modeled 74/80 (92.5%) of the ribosomal proteins, encompassing 12 archaeal/eukaryote-specific small subunit proteins as well as the complete complement of the ribosomal proteins of the eukaryotic large subunit. Near-complete atomic models of the 80S ribosome provide insights into the structure, function, and evolution of the eukaryotic translational apparatus.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Attenuation Correction for Whole Body PET/MR: Quantitative Evaluation and Lung Attenuation Estimation with Consistency Information

Bezrukov, I., Hofmann, M., Aschoff, P., Beyer, T., Mantlik, F., Pichler, B., Schölkopf, B.

2010(M13-122), 2010 Nuclear Science Symposium and Medical Imaging Conference (NSS-MIC), November 2010 (poster)

ei

[BibTex]

[BibTex]


no image
PET/MRI: Observation of Non-Isotropic Positron Distribution in High Magnetic Fields and Its Diagnostic Impact

Kolb, A., Hofmann, M., Sauter, A., Liu, C., Schölkopf, B., Pichler, B.

2010 Nuclear Science Symposium and Medical Imaging Conference, 2010(M18-119):1, November 2010 (poster)

ei

Web [BibTex]

Web [BibTex]


no image
Spatio-Spectral Remote Sensing Image Classification With Graph Kernels

Camps-Valls, G., Shervashidze, N., Borgwardt, K.

IEEE Geoscience and Remote Sensing Letters, 7(4):741-745, October 2010 (article)

Abstract
This letter presents a graph kernel for spatio-spectral remote sensing image classification with support vector machines (SVMs). The method considers higher order relations in the neighborhood (beyond pairwise spatial relations) to iteratively compute a kernel matrix for SVM learning. The proposed kernel is easy to compute and constitutes a powerful alternative to existing approaches. The capabilities of the method are illustrated in several multi- and hyperspectral remote sensing images acquired over both urban and agricultural areas.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Causal Inference Using the Algorithmic Markov Condition

Janzing, D., Schölkopf, B.

IEEE Transactions on Information Theory, 56(10):5168-5194, October 2010 (article)

Abstract
Inferring the causal structure that links $n$ observables is usually based upon detecting statistical dependences and choosing simple graphs that make the joint measure Markovian. Here we argue why causal inference is also possible when the sample size is one. We develop a theory how to generate causal graphs explaining similarities between single objects. To this end, we replace the notion of conditional stochastic independence in the causal Markov condition with the vanishing of conditional algorithmic mutual information and describe the corresponding causal inference rules. We explain why a consistent reformulation of causal inference in terms of algorithmic complexity implies a new inference principle that takes into account also the complexity of conditional probability densities, making it possible to select among Markov equivalent causal graphs. This insight provides a theoretical foundation of a heuristic principle proposed in earlier work. We also sketch some ideas on how to replace Kolmogorov complexity with decidable complexity criteria. This can be seen as an algorithmic analog of replacing the empirically undecidable question of statistical independence with practical independence tests that are based on implicit or explicit assumptions on the underlying distribution.

ei

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Recurrent Policy Gradients

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

Logic Journal of the IGPL, 18(5):620-634, October 2010 (article)

Abstract
Reinforcement learning for partially observable Markov decision problems (POMDPs) is a challenge as it requires policies with an internal state. Traditional approaches suffer significantly from this shortcoming and usually make strong assumptions on the problem domain such as perfect system models, state-estimators and a Markovian hidden system. Recurrent neural networks (RNNs) offer a natural framework for dealing with policy learning using hidden state and require only few limiting assumptions. As they can be trained well using gradient descent, they are suited for policy gradient approaches. In this paper, we present a policy gradient method, the Recurrent Policy Gradient which constitutes a model-free reinforcement learning method. It is aimed at training limited-memory stochastic policies on problems which require long-term memories of past observations. The approach involves approximating a policy gradient for a recurrent neural network by backpropagating return-weighted characteristic eligibilities through time. Using a ‘‘Long Short-Term Memory’’ RNN architecture, we are able to outperform previous RL methods on three important benchmark tasks. Furthermore, we show that using history-dependent baselines helps reducing estimation variance significantly, thus enabling our approach to tackle more challenging, highly stochastic environments.

ei

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Probabilistic Assignment of Chemical Shift Data for Semi-Automatic Amino Acid Recognition

Hooge, J.

11(10):30, 11th Conference of Junior Neuroscientists of T{\"u}bingen (NeNa), October 2010 (poster)

Abstract
manner. First the backbone resonances are assigned. This is usually achieved from sequential information provided by three chemical shifts: CA, CB and C’. Once the sequence is solved, the second assignment step takes place. For this purpose, the CA-CB and HA chemical shifts are used as a start point for assignment of the side chain resonances, thus connecting the backbone resonances to their respective side chains. This strategy is unfortunately limited by the size of the protein due to increasing signal overlap and missing signals. Therefore, amino acid recognition is in many cases not possible as the CA-CB chemical shift pattern is not sufficient to discriminate between the 20 amino acids. As a result, the first step of the strategy described above remains tedious and time consuming. The combination of modern NMR techniques with new spectrometers now provide information that was not always accessible in the past, due to sensitivity problems. These experiments can be applied efficiently to measure a protein size up to 45 kDa and furthermore provide a unique combination of sequential carbon spin system information. The assignment process can thus benefit from a maximum knowledge input, containing âallâ backbone and side chain chemical shifts as well as an immediate amino acid recognition from the side chain spin system. We propose to extend the software PASTA (Protein ASsignment by Threshold Accepting) to achieve a general sequential assignment of backbone and side-chain resonances in a semi- to fullautomatic per-residue approach. PASTA will offer the possibility to achieve the sequential assignment using any kind of chemical shifts (carbons and/or protons) that can provide sequential information combined with an amino acid recognition feature based on carbon spin system analysis.

ei

PDF [BibTex]

PDF [BibTex]


no image
Generalizing Demonstrated Actions in Manipulation Tasks

Kroemer, O., Detry, R., Piater, J., Peters, J.

IROS 2010 Workshop on Grasp Planning and Task Learning by Imitation, 2010, pages: 1, October 2010 (poster)

Abstract
Programming-by-demonstration promises to significantly reduce the burden of coding robots to perform new tasks. However, service robots will be presented with a variety of different situations that were not specifically demonstrated to it. In such cases, the robot must autonomously generalize its learned motions to these new situations. We propose a system that can generalize movements to new target locations and even new objects. The former is achieved by using a task-specific coordinate system together with dynamical systems motor primitives. Generalizing actions to new objects is a more complex problem, which we solve by treating it as a continuum-armed bandits problem. Using the bandits framework, we can efficiently optimize the learned action for a specific object. The proposed method was implemented on a real robot and succesfully adapted the grasping action to three different objects. Although we focus on grasping as an example of a task, the proposed methods are much more widely applicable to robot manipulation tasks.

ei

PDF Web [BibTex]

PDF Web [BibTex]


no image
Discriminative frequent subgraph mining with optimality guarantees

Thoma, M., Cheng, H., Gretton, A., Han, J., Kriegel, H., Smola, A., Song, L., Yu, P., Yan, X., Borgwardt, K.

Journal of Statistical Analysis and Data Mining, 3(5):302–318, October 2010 (article)

Abstract
The goal of frequent subgraph mining is to detect subgraphs that frequently occur in a dataset of graphs. In classification settings, one is often interested in discovering discriminative frequent subgraphs, whose presence or absence is indicative of the class membership of a graph. In this article, we propose an approach to feature selection on frequent subgraphs, called CORK, that combines two central advantages. First, it optimizes a submodular quality criterion, which means that we can yield a near-optimal solution using greedy feature selection. Second, our submodular quality function criterion can be integrated into gSpan, the state-of-the-art tool for frequent subgraph mining, and help to prune the search space for discriminative frequent subgraphs even during frequent subgraph mining.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Inhomogeneous Positron Range Effects in High Magnetic Fields might Cause Severe Artefacts in PET/MRI

Kolb, A., Hofmann, M., Sauter, A., Liu, C., Eriksson, L., Pichler, B.

(0305B), 2010 World Molecular Imaging Congress (WMIC), September 2010 (poster)

Abstract
The combination of PET and MRI is an emerging field of current research. It is known that the positron range is shortened in high magnetic fields (MF), leading to an improved resolution in PET images. Interestingly, only the fraction of positron range (PR) orthogonal to the MF is reduced and the fraction along the MF is not affected and yields to a non-isotropic count distribution. We measured the PR effect with PET isotopes like F-18, Cu-64, C-11, N-13 and Ga-68. A piece of paper (1 cm2) was soaked with each isotope and placed in the cFOV of a clinical 3T BrainPET/MR scanner. A polyethylene board (PE) was placed as a positron (β+) stopper with an axial distance of 3 cm from the soaked paper. The area under the peaks of one pixel wide profiles along the z-axis in coronal images was compared. Based on these measurements we confirmed our data in organic tissue. A larynx/trachea and lung of a butchered swine were injected with a mixture of NiSO4 for T1 MRI signals and Ga-68, simulating tumor lesions in the respiratory tract. The trachea/larynx were aligned in 35° to the MF lines and a small mass lesion was inserted to imitate a primary tracheal tumor whereas the larynx was injected submucosally in the lower medial part of the epiglottis. Reconstructed PET data show that the annihilated ratio of β+ at the origin position and in the PE depends on the isotope energy and the direction of the MF. The annihilation ratios of the source and PE are 52.4/47.6 (F-18), 57.5/42.5 (Cu-64), 43.7/56.7 (C-11), 31.1/68.9 (N-13) and 14.9/85.1 (Ga-68). In the swine larynx measurement, an artefact with approximately 39% of the lesion activity formed along MF lines 3cm away from the original injected position (fig.1). The data of the trachea showed two shine artefacts with a symmetric alignment along the MF lines. About 58% of the positrons annihilated at the lesion and 21% formed each artefact. The PR effects areminor in tissue of higher or equal density to water (0.096 cm-1). However, the effect is severe in low density tissue or air and might lead to misinterpretation of clinical data.

ei

Web [BibTex]

Web [BibTex]


no image
Combining active learning and reactive control for robot grasping

Kroemer, O., Detry, R., Piater, J., Peters, J.

Robotics and Autonomous Systems, 58(9):1105-1116, September 2010 (article)

Abstract
Grasping an object is a task that inherently needs to be treated in a hybrid fashion. The system must decide both where and how to grasp the object. While selecting where to grasp requires learning about the object as a whole, the execution only needs to reactively adapt to the context close to the grasp’s location. We propose a hierarchical controller that reflects the structure of these two sub-problems, and attempts to learn solutions that work for both. A hybrid architecture is employed by the controller to make use of various machine learning methods that can cope with the large amount of uncertainty inherent to the task. The controller’s upper level selects where to grasp the object using a reinforcement learner, while the lower level comprises an imitation learner and a vision-based reactive controller to determine appropriate grasping motions. The resulting system is able to quickly learn good grasps of a novel object in an unstructured environment, by executing smooth reaching motions and preshapin g the hand depending on the object’s geometry. The system was evaluated both in simulation and on a real robot.

ei

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Nonparametric Regression between General Riemannian Manifolds

Steinke, F., Hein, M., Schölkopf, B.

SIAM Journal on Imaging Sciences, 3(3):527-563, September 2010 (article)

Abstract
We study nonparametric regression between Riemannian manifolds based on regularized empirical risk minimization. Regularization functionals for mappings between manifolds should respect the geometry of input and output manifold and be independent of the chosen parametrization of the manifolds. We define and analyze the three most simple regularization functionals with these properties and present a rather general scheme for solving the resulting optimization problem. As application examples we discuss interpolation on the sphere, fingerprint processing, and correspondence computations between three-dimensional surfaces. We conclude with characterizing interesting and sometimes counterintuitive implications and new open problems that are specific to learning between Riemannian manifolds and are not encountered in multivariate regression in Euclidean space.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Hybrid PET/MRI of Intracranial Masses: Initial Experiences and Comparison to PET/CT

Boss, A., Bisdas, S., Kolb, A., Hofmann, M., Ernemann, U., Claussen, C., Pfannenberg, C., Pichler, B., Reimold, M., Stegger, L.

Journal of Nuclear Medicine, 51(8):1198-1205, August 2010 (article)

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
libDAI: A Free and Open Source C++ Library for Discrete Approximate Inference in Graphical Models

Mooij, JM.

Journal of Machine Learning Research, 11, pages: 2169-2173, August 2010 (article)

Abstract
This paper describes the software package libDAI, a free & open source C++ library that provides implementations of various exact and approximate inference methods for graphical models with discrete-valued variables. libDAI supports directed graphical models (Bayesian networks) as well as undirected ones (Markov random fields and factor graphs). It offers various approximations of the partition sum, marginal probability distributions and maximum probability states. Parameter learning is also supported. A feature comparison with other open source software packages for approximate inference is given. libDAI is licensed under the GPL v2+ license and is available at http://www.libdai.org.

ei

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
Convolutive blind source separation by efficient blind deconvolution and minimal filter distortion

Zhang, K., Chan, L.

Neurocomputing, 73(13-15):2580-2588, August 2010 (article)

Abstract
Convolutive blind source separation (BSS) usually encounters two difficulties—the filter indeterminacy in the recovered sources and the relatively high computational load. In this paper we propose an efficient method to convolutive BSS, by dealing with these two issues. It consists of two stages, namely, multichannel blind deconvolution (MBD) and learning the post-filters with the minimum filter distortion (MFD) principle. We present a computationally efficient approach to MBD in the first stage: a vector autoregression (VAR) model is first fitted to the data, admitting a closed-form solution and giving temporally independent errors; traditional independent component analysis (ICA) is then applied to these errors to produce the MBD results. In the second stage, the least linear reconstruction error (LLRE) constraint of the separation system, which was previously used to regularize the solutions to nonlinear ICA, enforces a MFD principle of the estimated mixing system for convolutive BSS. One can then easily learn the post-filters to preserve the temporal structure of the sources. We show that with this principle, each recovered source is approximately the principal component of the contributions of this source to all observations. Experimental results on both synthetic data and real room recordings show the good performance of this method.

ei

PDF PDF DOI [BibTex]


no image
Biased Feedback in Brain-Computer Interfaces

Barbero, A., Grosse-Wentrup, M.

Journal of NeuroEngineering and Rehabilitation, 7(34):1-4, July 2010 (article)

Abstract
Even though feedback is considered to play an important role in learning how to operate a brain-computer interface (BCI), to date no significant influence of feedback design on BCI-performance has been reported in literature. In this work, we adapt a standard motor-imagery BCI-paradigm to study how BCI-performance is affected by biasing the belief subjects have on their level of control over the BCI system. Our findings indicate that subjects already capable of operating a BCI are impeded by inaccurate feedback, while subjects normally performing on or close to chance level may actually benefit from an incorrect belief on their performance level. Our results imply that optimal feedback design in BCIs should take into account a subject‘s current skill level.

ei

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Varieties of Justification in Machine Learning

Corfield, D.

Minds and Machines, 20(2):291-301, July 2010 (article)

Abstract
Forms of justification for inductive machine learning techniques are discussed and classified into four types. This is done with a view to introduce some of these techniques and their justificatory guarantees to the attention of philosophers, and to initiate a discussion as to whether they must be treated separately or rather can be viewed consistently from within a single framework.

ei

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Dirichlet Process Gaussian Mixture Models: Choice of the Base Distribution

Görür, D., Rasmussen, C.

Journal of Computer Science and Technology, 25(4):653-664, July 2010 (article)

Abstract
In the Bayesian mixture modeling framework it is possible to infer the necessary number of components to model the data and therefore it is unnecessary to explicitly restrict the number of components. Nonparametric mixture models sidestep the problem of finding the “correct” number of mixture components by assuming infinitely many components. In this paper Dirichlet process mixture (DPM) models are cast as infinite mixture models and inference using Markov chain Monte Carlo is described. The specification of the priors on the model parameters is often guided by mathematical and practical convenience. The primary goal of this paper is to compare the choice of conjugate and non-conjugate base distributions on a particular class of DPM models which is widely used in applications, the Dirichlet process Gaussian mixture model (DPGMM). We compare computational efficiency and modeling performance of DPGMM defined using a conjugate and a conditionally conjugate base distribution. We show that better density models can result from using a wider class of priors with no or only a modest increase in computational effort.

ei

PDF PDF DOI [BibTex]

PDF PDF DOI [BibTex]


no image
Robust probabilistic superposition and comparison of protein structures

Mechelke, M., Habeck, M.

BMC Bioinformatics, 11(363):1-13, July 2010 (article)

ei

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
Reinforcement Learning by Relative Entropy Policy Search

Peters, J., Mülling, K., Altun, Y.

30th International Workshop on Bayesian Inference and Maximum Entropy Methods in Science and Engineering (MaxEnt 2010), 30, pages: 69, July 2010 (poster)

Abstract
Policy search is a successful approach to reinforcement learning. However, policy improvements often result in the loss of information. Hence, it has been marred by premature convergence and implausible solutions. As first suggested in the context of covariant policy gradients, many of these problems may be addressed by constraining the information loss. In this book chapter, we continue this path of reasoning and suggest the Relative Entropy Policy Search (REPS) method. The resulting method differs significantly from previous policy gradient approaches and yields an exact update step. It works well on typical reinforcement learning benchmark problems. We will also present a real-world applications where a robot employs REPS to learn how to return balls in a game of table tennis.

ei

PDF [BibTex]

PDF [BibTex]


no image
Results of the GREAT08 Challenge: An image analysis competition for cosmological lensing

Bridle, S., Balan, S., Bethge, M., Gentile, M., Harmeling, S., Heymans, C., Hirsch, M., Hosseini, R., Jarvis, M., Kirk, D., Kitching, T., Kuijken, K., Lewis, A., Paulin-Henriksson, S., Schölkopf, B., Velander, M., Voigt, L., Witherick, D., Amara, A., Bernstein, G., Courbin, F., Gill, M., Heavens, A., Mandelbaum, R., Massey, R., Moghaddam, B., Rassat, A., Refregier, A., Rhodes, J., Schrabback, T., Shawe-Taylor, J., Shmakova, M., van Waerbeke, L., Wittman, D.

Monthly Notices of the Royal Astronomical Society, 405(3):2044-2061, July 2010 (article)

Abstract
We present the results of the GREAT08 Challenge, a blind analysis challenge to infer weak gravitational lensing shear distortions from images. The primary goal was to stimulate new ideas by presenting the problem to researchers outside the shear measurement community. Six GREAT08 Team methods were presented at the launch of the Challenge and five additional groups submitted results during the 6 month competition. Participants analyzed 30 million simulated galaxies with a range in signal to noise ratio, point-spread function ellipticity, galaxy size, and galaxy type. The large quantity of simulations allowed shear measurement methods to be assessed at a level of accuracy suitable for currently planned future cosmic shear observations for the first time. Different methods perform well in different parts of simulation parameter space and come close to the target level of accuracy in several of these. A number of fresh ideas have emerged as a result of the Challenge including a re-examination of the process of combining information from different galaxies, which reduces the dependence on realistic galaxy modelling. The image simulations will become increasingly sophis- ticated in future GREAT challenges, meanwhile the GREAT08 simulations remain as a benchmark for additional developments in shear measurement algorithms.

ei

Web DOI [BibTex]

Web DOI [BibTex]


no image
Remote Sensing Feature Selection by Kernel Dependence Estimation

Camps-Valls, G., Mooij, J., Schölkopf, B.

IEEE Geoscience and Remote Sensing Letters, 7(3):587-591, July 2010 (article)

Abstract
This letter introduces a nonlinear measure of independence between random variables for remote sensing supervised feature selection. The so-called Hilbert–Schmidt independence criterion (HSIC) is a kernel method for evaluating statistical dependence and it is based on computing the Hilbert–Schmidt norm of the cross-covariance operator of mapped samples in the corresponding Hilbert spaces. The HSIC empirical estimator is easy to compute and has good theoretical and practical properties. Rather than using this estimate for maximizing the dependence between the selected features and the class labels, we propose the more sensitive criterion of minimizing the associated HSIC p-value. Results in multispectral, hyperspectral, and SAR data feature selection for classification show the good performance of the proposed approach.

ei

PDF Web DOI [BibTex]

PDF Web DOI [BibTex]


no image
Clustering stability: an overview

von Luxburg, U.

Foundations and Trends in Machine Learning, 2(3):235-274, July 2010 (article)

Abstract
A popular method for selecting the number of clusters is based on stability arguments: one chooses the number of clusters such that the corresponding clustering results are "most stable". In recent years, a series of papers has analyzed the behavior of this method from a theoretical point of view. However, the results are very technical and difficult to interpret for non-experts. In this paper we give a high-level overview about the existing literature on clustering stability. In addition to presenting the results in a slightly informal but accessible way, we relate them to each other and discuss their different implications.

ei

PDF DOI [BibTex]

PDF DOI [BibTex]


no image
A Maximum Entropy Approach to Semi-supervised Learning

Erkan, A., Altun, Y.

30th International Workshop on Bayesian Inference and Maximum Entropy Methods in Science and Engineering (MaxEnt 2010), 30, pages: 80, July 2010 (poster)

Abstract
Maximum entropy (MaxEnt) framework has been studied extensively in supervised learning. Here, the goal is to find a distribution p that maximizes an entropy function while enforcing data constraints so that the expected values of some (pre-defined) features with respect to p match their empirical counterparts approximately. Using different entropy measures, different model spaces for p and different approximation criteria for the data constraints yields a family of discriminative supervised learning methods (e.g., logistic regression, conditional random fields, least squares and boosting). This framework is known as the generalized maximum entropy framework. Semi-supervised learning (SSL) has emerged in the last decade as a promising field that combines unlabeled data along with labeled data so as to increase the accuracy and robustness of inference algorithms. However, most SSL algorithms to date have had trade-offs, e.g., in terms of scalability or applicability to multi-categorical data. We extend the generalized MaxEnt framework to develop a family of novel SSL algorithms. Extensive empirical evaluation on benchmark data sets that are widely used in the literature demonstrates the validity and competitiveness of the proposed algorithms.

ei

PDF PDF [BibTex]

PDF PDF [BibTex]


no image
The effect of positioning aids on PET quantification following MR-based attenuation correction (AC) in PET/MR imaging

Mantlik, F., Hofmann, M., Kupferschläger, J., Werner, M., Pichler, B., Beyer, T.

Journal of Nuclear Medicine, 51(Supplement 2):1418 , June 2010 (poster)

Abstract
Objectives: We study the quantitative effect of not accounting for the attenuation of patient positioning aids in combined PET/MR imaging. Methods: Positioning aids cannot be detected with conventional MR sequences. We mimic this effect using PET/CT data (Biograph HiRez16) with the foams removed from CT images prior to using them for CT-AC. PET/CT data were acquired using standard parameters (phantoms/patients): 120/140 kVp, 30/250 mAs, 5 mm slices, OSEM (4i, 8s, 5 mm filter) following CT-AC. First, a uniform 68Ge-cylinder was positioned centrally in the PET/CT and fixed with a vacuum mattress (10 cm thick). Second, the same cylinder was placed in 3 positioning aids from the PET/MR (BrainPET-3T). Third, 5 head/neck patients who were fixed in a vacuum mattress were selected. In all 3 studies PET recon post CT-AC based on measured CT images was used as the reference (mCT-AC). The PET/MR set-up was mimicked by segmenting the foam inserts from the measured CT images and setting their voxel values to -1000 HU (air). PET images were reconstructed using CT-AC with the segmented CT images (sCT-AC). PET images with mCT- and sCT-AC were compared. Results: sCT-AC underestimated PET voxel values in the phantom by 6.7% on average compared to mCT-AC with the vacuum mattress in place. 5% of the PET voxels were underestimated by >=10%. Not accounting for MR positioning aids during AC led to an underestimation of 2.8% following sCT-AC, with 5% of the PET voxels being underestimated by >=7% wrt mCT-AC. Preliminary evaluation of the patient data indicates a slightly higher bias from not accounting for patient positioning aids (mean: -9.1%, 5% percentile: -11.2%). Conclusions: A considerable and regionally variable underestimation of the PET activity following AC is observed when positioning aids are not accounted for. This bias may become relevant in neurological activation or dementia studies with PET/MR

ei

Web [BibTex]

Web [BibTex]


no image
Justifying Additive Noise Model-Based Causal Discovery via Algorithmic Information Theory

Janzing, D., Steudel, B.

Open Systems and Information Dynamics, 17(2):189-212, June 2010 (article)

Abstract
A recent method for causal discovery is in many cases able to infer whether X causes Y or Y causes X for just two observed variables X and Y. It is based on the observation that there exist (non-Gaussian) joint distributions P(X,Y) for which Y may be written as a function of X up to an additive noise term that is independent of X and no such model exists from Y to X. Whenever this is the case, one prefers the causal model X → Y. Here we justify this method by showing that the causal hypothesis Y → X is unlikely because it requires a specific tuning between P(Y) and P(X|Y) to generate a distribution that admits an additive noise model from X to Y. To quantify the amount of tuning, needed we derive lower bounds on the algorithmic information shared by P(Y) and P(X|Y). This way, our justification is consistent with recent approaches for using algorithmic information theory for causal reasoning. We extend this principle to the case where P(X,Y) almost admits an additive noise model. Our results suggest that the above conclusion is more reliable if the complexity of P(Y) is high.

ei

PDF Web DOI [BibTex]


no image
Multi-task Learning for Zero Training Brain-Computer Interfaces

Alamgir, M., Grosse-Wentrup, M., Altun, Y.

4th International BCI Meeting, June 2010 (poster)

Abstract
Brain-computer interfaces (BCIs) are limited in their applicability in everyday settings by the current necessity to record subject-specific calibration data prior to actual use of the BCI for communication. In this work, we utilize the framework of multitask learning to construct a BCI that can be used without any subject-specific calibration process, i.e., with zero training data. In BCIs based on EEG or MEG, the predictive function of a subject's intention is commonly modeled as a linear combination of some features derived from spatial and spectral recordings. The coefficients of this combination correspond to the importance of the features for predicting the intention of the subject. These coefficients are usually learned separately for each subject due to inter-subject variability. Principle feature characteristics, however, are known to remain invariant across subject. For example, it is well known that in motor imagery paradigms spectral power in the mu- and beta frequency ranges (roughly 8-14 Hz and 20-30 Hz, respectively) over sensorimotor areas provides most information on a subject's intention. Based on this assumption, we define the intention prediction function as a combination of subject-invariant and subject-specific models, and propose a machine learning method that infers these models jointly using data from multiple subjects. This framework leads to an out-of-the-box intention predictor, where the subject-invariant model can be employed immediately for a subject with no prior data. We present a computationally efficient method to further improve this BCI to incorporate subject-specific variations as such data becomes available. To overcome the problem of high dimensional feature spaces in this context, we further present a new method for finding the relevance of different recording channels according to actions performed by subjects. Usually, the BCI feature representation is a concatenation of spectral features extracted from different channels. This representation, however, is redundant, as recording channels at different spatial locations typically measure overlapping sources within the brain due to volume conduction. We address this problem by assuming that the relevance of different spectral bands is invariant across channels, while learning different weights for each recording electrode. This framework allows us to significantly reduce the feature space dimensionality without discarding potentially useful information. Furthermore, the resulting out-of-the-box BCI can be adapted to different experimental setups, for example EEG caps with different numbers of channels, as long as there exists a mapping across channels in different setups. We demonstrate the feasibility of our approach on a set of experimental EEG data recorded during a standard two-class motor imagery paradigm from a total of ten healthy subjects. Specifically, we show that satisfactory classification results can be achieved with zero training data, and that combining prior recordings with subject-specific calibration data substantially outperforms using subject-specific data only.

ei

Web [BibTex]


no image
Causal Influence of Gamma Oscillations on Performance in Brain-Computer Interfaces

Grosse-Wentrup, M., Hill, J., Schölkopf, B.

4th International BCI Meeting0, June 2010 (poster)

Abstract
Background and Objective: While machine learning approaches have led to tremendous advances in brain-computer interfaces (BCIs) in recent years (cf. [1]), there still exists a large variation in performance across subjects. Furthermore, a significant proportion of subjects appears incapable of achieving above chance-level classification accuracy [2], which to date includes all subjects in a completely locked-in state that have been trained in BCI control. Understanding the reasons for this variation in performance arguably constitutes one of the most fundamental open questions in research on BCIs. Methods & Results Using a machine learning approach, we derive a trial-wise measure of how well EEG recordings can be classified as either left- or right-hand motor imagery. Specifically, we train a support vector machine (SVM) on log-bandpower features (7-40 Hz) derived from EEG channels after spatial filtering with a surface Laplacian, and then compute the trial-wise distance of the output of the SVM from the separating hyperplane using a cross-validation procedure. We then correlate this trial-wise performance measure, computed on EEG recordings of ten healthy subjects, with log-bandpower in the gamma frequency range (55-85 Hz), and demonstrate that it is positively correlated with frontal- and occipital gamma-power and negatively correlated with centro-parietal gamma-power. This correlation is shown to be highly significant on the group level as well as in six out of ten subjects on the single-subject level. We then utilize the framework for causal inference developed by Pearl, Spirtes and others [3,4] to present evidence that gamma-power is not only correlated with BCI performance but does indeed exert a causal influence on it. Discussion and Conclusions Our results indicate that successful execution of motor imagery, and hence reliable communication by means of a BCI based on motor imagery, requires a volitional shift of gamma-power from centro-parietal to frontal and occipital regions. As such, our results provide the first non-trivial explanation for the variation in BCI performance across and within subjects. As this topographical alteration in gamma-power is likely to correspond to a specific attentional shift, we propose to provide subjects with feedback on their topographical distribution of gamma-power in order to establish the attentional state required for successful execution of motor imagery.

ei

Web [BibTex]