Publications | Max Planck Institute for Intelligent Systems

210 results (View BibTeX file of all listed publications)

2013

Correlation of Simultaneously Acquired Diffusion-Weighted Imaging and 2-Deoxy-[18F] fluoro-2-D-glucose Positron Emission Tomography of Pulmonary Lesions in a Dedicated Whole-Body Magnetic Resonance/Positron Emission Tomography System

Schmidt, H., Brendle, C., Schraml, C., Martirosian, P., Bezrukov, I., Hetzel, J., Müller, M., Sauter, A., Claussen, C., Pfannenberg, C., Schwenzer, N.

Investigative Radiology, 48(5):247-255, May 2013 (article)

Web [BibTex]

2013

ei Schmidt, H., Brendle, C., Schraml, C., Martirosian, P., Bezrukov, I., Hetzel, J., Müller, M., Sauter, A., Claussen, C., Pfannenberg, C., Schwenzer, N. Correlation of Simultaneously Acquired Diffusion-Weighted Imaging and 2-Deoxy-[18F] fluoro-2-D-glucose Positron Emission Tomography of Pulmonary Lesions in a Dedicated Whole-Body Magnetic Resonance/Positron Emission Tomography System Investigative Radiology, 48(5):247-255, May 2013 (article)

Web [BibTex]

Replacing Causal Faithfulness with Algorithmic Independence of Conditionals

Lemeire, J., Janzing, D.

Minds and Machines, 23(2):227-249, May 2013 (article)

Abstract

Independence of Conditionals (IC) has recently been proposed as a basic rule for causal structure learning. If a Bayesian network represents the causal structure, its Conditional Probability Distributions (CPDs) should be algorithmically independent. In this paper we compare IC with causal faithfulness (FF), stating that only those conditional independences that are implied by the causal Markov condition hold true. The latter is a basic postulate in common approaches to causal structure learning. The common spirit of FF and IC is to reject causal graphs for which the joint distribution looks ‘non-generic’. The difference lies in the notion of genericity: FF sometimes rejects models just because one of the CPDs is simple, for instance if the CPD describes a deterministic relation. IC does not behave in this undesirable way. It only rejects a model when there is a non-generic relation between different CPDs although each CPD looks generic when considered separately. Moreover, it detects relations between CPDs that cannot be captured by conditional independences. IC therefore helps in distinguishing causal graphs that induce the same conditional independences (i.e., they belong to the same Markov equivalence class). The usual justification for FF implicitly assumes a prior that is a probability density on the parameter space. IC can be justified by Solomonoff’s universal prior, assigning non-zero probability to those points in parameter space that have a finite description. In this way, it favours simple CPDs, and therefore respects Occam’s razor. Since Kolmogorov complexity is uncomputable, IC is not directly applicable in practice. We argue that it is nevertheless helpful, since it has already served as inspiration and justification for novel causal inference algorithms.

PDF Web DOI [BibTex]

ei Lemeire, J., Janzing, D. Replacing Causal Faithfulness with Algorithmic Independence of Conditionals Minds and Machines, 23(2):227-249, May 2013 (article)

PDF Web DOI [BibTex]

What can neurons do for their brain? Communicate selectivity with bursts

Balduzzi, D., Tononi, G.

Theory in Biosciences , 132(1):27-39, Springer, March 2013 (article)

Abstract

Neurons deep in cortex interact with the environment extremely indirectly; the spikes they receive and produce are pre- and post-processed by millions of other neurons. This paper proposes two information-theoretic constraints guiding the production of spikes, that help ensure bursting activity deep in cortex relates meaningfully to events in the environment. First, neurons should emphasize selective responses with bursts. Second, neurons should propagate selective inputs by burst-firing in response to them. We show the constraints are necessary for bursts to dominate information-transfer within cortex, thereby providing a substrate allowing neurons to distribute credit amongst themselves. Finally, since synaptic plasticity degrades the ability of neurons to burst selectively, we argue that homeostatic regulation of synaptic weights is necessary, and that it is best performed offline during sleep.

PDF Web DOI [BibTex]

ei Balduzzi, D., Tononi, G. What can neurons do for their brain? Communicate selectivity with bursts Theory in Biosciences , 132(1):27-39, Springer, March 2013 (article)

PDF Web DOI [BibTex]

Apprenticeship Learning with Few Examples

Boularias, A., Chaib-draa, B.

Neurocomputing, 104, pages: 83-96, March 2013 (article)

Abstract

We consider the problem of imitation learning when the examples, provided by an expert human, are scarce. Apprenticeship learning via inverse reinforcement learning provides an efficient tool for generalizing the examples, based on the assumption that the expert's policy maximizes a value function, which is a linear combination of state and action features. Most apprenticeship learning algorithms use only simple empirical averages of the features in the demonstrations as a statistics of the expert's policy. However, this method is efficient only when the number of examples is sufficiently large to cover most of the states, or the dynamics of the system is nearly deterministic. In this paper, we show that the quality of the learned policies is sensitive to the error in estimating the averages of the features when the dynamics of the system is stochastic. To reduce this error, we introduce two new approaches for bootstrapping the demonstrations by assuming that the expert is near-optimal and the dynamics of the system is known. In the first approach, the expert's examples are used to learn a reward function and to generate furthermore examples from the corresponding optimal policy. The second approach uses a transfer technique, known as graph homomorphism, in order to generalize the expert's actions to unvisited regions of the state space. Empirical results on simulated robot navigation problems show that our approach is able to learn sufficiently good policies from a significantly small number of examples.

Web DOI [BibTex]

ei Boularias, A., Chaib-draa, B. Apprenticeship Learning with Few Examples Neurocomputing, 104, pages: 83-96, March 2013 (article)

Web DOI [BibTex]

Quasi-Newton Methods: A New Direction

Hennig, P., Kiefel, M.

Journal of Machine Learning Research, 14(1):843-865, March 2013 (article)

Abstract

Four decades after their invention, quasi-Newton methods are still state of the art in unconstrained numerical optimization. Although not usually interpreted thus, these are learning algorithms that fit a local quadratic approximation to the objective function. We show that many, including the most popular, quasi-Newton methods can be interpreted as approximations of Bayesian linear regression under varying prior assumptions. This new notion elucidates some shortcomings of classical algorithms, and lights the way to a novel nonparametric quasi-Newton method, which is able to make more efficient use of available information at computational cost similar to its predecessors.

ei ps pn

website+code pdf link (url) [BibTex]

ei ps pn Hennig, P., Kiefel, M. Quasi-Newton Methods: A New Direction Journal of Machine Learning Research, 14(1):843-865, March 2013 (article)

website+code pdf link (url) [BibTex]

Regional effects of magnetization dispersion on quantitative perfusion imaging for pulsed and continuous arterial spin labeling

Cavusoglu, M., Pohmann, R., Burger, H. C., Uludag, K.

Magnetic Resonance in Medicine, 69(2):524-530, Febuary 2013 (article)

Abstract

Most experiments assume a global transit delay time with blood flowing from the tagging region to the imaging slice in plug flow without any dispersion of the magnetization. However, because of cardiac pulsation, nonuniform cross-sectional flow profile, and complex vessel networks, the transit delay time is not a single value but follows a distribution. In this study, we explored the regional effects of magnetization dispersion on quantitative perfusion imaging for varying transit times within a very large interval from the direct comparison of pulsed, pseudo-continuous, and dual-coil continuous arterial spin labeling encoding schemes. Longer distances between tagging and imaging region typically used for continuous tagging schemes enhance the regional bias on the quantitative cerebral blood flow measurement causing an underestimation up to 37% when plug flow is assumed as in the standard model.

Web DOI [BibTex]

ei Cavusoglu, M., Pohmann, R., Burger, H. C., Uludag, K. Regional effects of magnetization dispersion on quantitative perfusion imaging for pulsed and continuous arterial spin labeling Magnetic Resonance in Medicine, 69(2):524-530, Febuary 2013 (article)

Web DOI [BibTex]

The multivariate Watson distribution: Maximum-likelihood estimation and other aspects

Sra, S., Karp, D.

Journal of Multivariate Analysis, 114, pages: 256-269, February 2013 (article)

Abstract

This paper studies fundamental aspects of modelling data using multivariate Watson distributions. Although these distributions are natural for modelling axially symmetric data (i.e., unit vectors where View the MathML source are equivalent), for high-dimensions using them can be difficult—largely because for Watson distributions even basic tasks such as maximum-likelihood are numerically challenging. To tackle the numerical difficulties some approximations have been derived. But these are either grossly inaccurate in high-dimensions [K.V. Mardia, P. Jupp, Directional Statistics, second ed., John Wiley & Sons, 2000] or when reasonably accurate [A. Bijral, M. Breitenbach, G.Z. Grudic, Mixture of Watson distributions: a generative model for hyperspherical embeddings, in: Artificial Intelligence and Statistics, AISTATS 2007, 2007, pp. 35–42], they lack theoretical justification. We derive new approximations to the maximum-likelihood estimates; our approximations are theoretically well-defined, numerically accurate, and easy to compute. We build on our parameter estimation and discuss mixture-modelling with Watson distributions; here we uncover a hitherto unknown connection to the “diametrical clustering” algorithm of Dhillon et al. [I.S. Dhillon, E.M. Marcotte, U. Roshan, Diametrical clustering for identifying anticorrelated gene clusters, Bioinformatics 19 (13) (2003) 1612–1619].

Web DOI [BibTex]

ei Sra, S., Karp, D. The multivariate Watson distribution: Maximum-likelihood estimation and other aspects Journal of Multivariate Analysis, 114, pages: 256-269, February 2013 (article)

Web DOI [BibTex]

How the result of graph clustering methods depends on the construction of the graph

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

ESAIM: Probability & Statistics, 17, pages: 370-418, January 2013 (article)

Abstract

We study the scenario of graph-based clustering algorithms such as spectral clustering. Given a set of data points, one rst has to construct a graph on the data points and then apply a graph clustering algorithm to nd a suitable partition of the graph. Our main question is if and how the construction of the graph (choice of the graph, choice of parameters, choice of weights) in uences the outcome of the nal clustering result. To this end we study the convergence of cluster quality measures such as the normalized cut or the Cheeger cut on various kinds of random geometric graphs as the sample size tends to innity. It turns out that the limit values of the same objective function are systematically dierent on dierent types of graphs. This implies that clustering results systematically depend on the graph and can be very dierent for dierent types of graph. We provide examples to illustrate the implications on spectral clustering.

PDF DOI [BibTex]

ei Maier, M., von Luxburg, U., Hein, M. How the result of graph clustering methods depends on the construction of the graph ESAIM: Probability & Statistics, 17, pages: 370-418, January 2013 (article)

PDF DOI [BibTex]

Falsification and future performance

Balduzzi, D.

In Algorithmic Probability and Friends. Bayesian Prediction and Artificial Intelligence, 7070, pages: 65-78, Lecture Notes in Computer Science, Springer, Berlin, Germany, Solomonoff 85th Memorial Conference, January 2013 (inproceedings)

Abstract

We information-theoretically reformulate two measures of capacity from statistical learning theory: empirical VC-entropy and empirical Rademacher complexity. We show these capacity measures count the number of hypotheses about a dataset that a learning algorithm falsifies when it finds the classifier in its repertoire minimizing empirical risk. It then follows from that the future performance of predictors on unseen data is controlled in part by how many hypotheses the learner falsifies. As a corollary we show that empirical VC-entropy quantifies the message length of the true hypothesis in the optimal code of a particular probability distribution, the so-called actual repertoire.

PDF Web DOI [BibTex]

ei Balduzzi, D. Falsification and future performance In Algorithmic Probability and Friends. Bayesian Prediction and Artificial Intelligence, 7070, pages: 65-78, Lecture Notes in Computer Science, Springer, Berlin, Germany, Solomonoff 85th Memorial Conference, January 2013 (inproceedings)

PDF Web DOI [BibTex]

Explicit eigenvalues of certain scaled trigonometric matrices

Sra, S.

Linear Algebra and its Applications, 438(1):173-181, January 2013 (article)

MPI Papers

Departments

Research Groups

Publication Type

Year

2013

2013

Latest News

Links

Contact Us