## Title and Abstract of talks:

### March 13rd (Thu)

Michael Elad (Technion, Israel)

Sparse Modeling of Graph-Structured Data ... and ... Images.

slides: ppt,
pdf

(*) This talk covers portions from a joint work with Idan Ram and Israel Cohen.

Yoshiyuki Kabashima (Tokyo Institute of Technology)

Statistical mechanical approach to sample complexity of dictionary learning

#This talk is based on a collaboration with Ayaka Sakata.

Toshiyuki Tanaka (Kyoto University)

TBA

Shiro Ikeda (Institute of Statistical Mathematics)

Some applications of sparse modelling in physical measurements

slides

Kenji Nagata (University of Tokyo)

Sparse modeling and variable selection with the exchange Monte Carlo method

### March 14th (Fri)

Noureddine El Karoui (UC Berkeley, USA)

On high-dimensional robust regression, penalized robust regression and GLMs

Many surprising statistical phenomena occur: for instance, maximum likelihood methods are shown to be (grossly) inefficient, and loss functions that should be used in regression are shown to depend on the ratio p/n. This means that dimensionality should be explicitly taken into account when performing simple tasks such as regression.

More generally, we'll see that intuition based on results obtained in the small p, large n setting leads to misconceptions and the use of suboptimal procedures. It also turns out that inference is possible in this setting. We'll also see that the geometry of the design matrix plays a key role in these problems and use this fact to disprove claims of universality of some of the results.

Mathematically, the tools needed mainly come from random matrix theory, measure concentration and convex analysis.

Kei Kobayashi (Institute of Statistical Mathematics)

Curvature of empirical metrics on a data space and its deformation

This is a collaboration with Prof. Henry P. Wynn (London School of Economics).

Kengo Kato (University of Tokyo)

Gaussian approximations and multiplier bootstrap for maxima of sums of
high-dimensional random vectors

(joint work with V. Chernozhukov and D. Chetverikov)

slides

Arthur Gretton (University College London)

Kernel Distribution Embeddings and Tests for Three-Variable Interactions

slides

The Lancaster and total independence statistics are straightforward to compute, and the resulting tests are are consistent against all alternatives for a large family of reproducing kernels. We show the Lancaster test to be sensitive to cases where two independent causes individually have weak influence on a third dependent variable, but their combined effect has a strong influence. This makes the Lancaster test especially suited to finding structure in directed graphical models, where it outperforms competing nonparametric tests in detecting such V-structures.

Relevant paper: Sejdinovic, D., Gretton, A., and Bergsma, W., A Kernel Test for Three-Variable Interactions, NIPS, 2013

Shotaro Akaho
(Advanced Institute of Science and Technology)

On the robust nonlinear curve fitting

slides

Kenji Fukumizu
(Institute of Statistical Mathematics)

Nonparametric Bayesian Inference with Positive Definite Kernels

slides

### March 15th (Sat)

Milos Radovanovic
@(Univ. Novi Sad, Serbia)

Hubs in Nearest-Neighbor Graphs: Origins, Applications and Challenges

slides

Ikumi Suzuki
(National Institute of Genetics)

The effect of data centering for k-nearest neighbor

slides

Lately, as one of the of curse of dimensionality problem, hubness phenomena is reported in terms of the kNN. A "hub" is a sample which is similar to many other samples in a dataset.

While spectral property of the laplacian-based kernels is useful for the dimensionality reduction and the spectral clustering etc., this property also effects on the kNN, namely for reduction of hub samples.

On further study, simply centring similarity measures, which can be seen as shifting the data origin to the data centroid in a feature space (data centering), has also property of hub reduction.

Ryota Tomioka
(Toyota Technological Institute, Chicago)

Towards better computation-statistics trade-off in tensor decomposition

slides

Taiji Suzuki
(Tokyo Institute of Technology)

PAC-Bayesian Bound for Gaussian Process Regression and Multiple Kernel Additive Model

slides