Forschungsgruppe "Stochastische Algorithmen und Nichtparametrische Statistik"

15.04.13 (Monday!)

Name: Dr. Andrea Fuster

Titel: Riemannian framework for brain diffusion MRI

Abstract: One of the approaches in the analysis of brain diffusion MRI data is to consider white matter as a Riemannian manifold, with a metric related to the diffusion tensor. In this framework geodesic tractography can be performed to infer the architecture of white matter pathways in the brain. We propose a novel Riemannian metric and present promising tractography results.

23.04.2013

No Talk!

30.04.2013

Dr. Alexander Goldenshluger

A unified framework for change point detection and other related problems

Abstract: We propose a unified convex-optimization-based framework for problems of detecting a signal of a given shape in Gaussian noise. The framework covers various detection settings including: detection of jumps in curves and their derivatives; detection of a periodic component in Gaussian time series; and signal detection from indirect observations. We present a general detection procedure, analyze its properties and show that it cannot be improved in some specific settings.

07.05.2013

Marcel Ladkau (WIAS Berlin)

A new multi-factor stochastic volatility model with displacement

Abstract: tba

14.05.2013

Mathias Trabs (HU Berlin)

Semiparametric efficiency for inverse problems with applications to deconvolution and Lévy models

Abstract: Starting with a linear inverse problem in an abstract white noise model, we study semiparametric efficiency in the sense of Hájek-Le Cam. The inverse problem embeds into a general setup where a convolution theorem is obtained. The local behavior of the model is described by the score operator. The results are then applied to a deconvolution setting and to the nonlinear inverse problem in a Lévy model. In the latter we observe a Lévy process at low frequency and we determine the information bound for linear functionals of the jump density. Of particular interest is the estimation of the distribution function in both settings. Locally the deconvolution model and the Lévy model converge to limit experiments which are given by a white noise inverse problem with an appropriate linear operator. The lower bounds coincide with recent results on the asymptotic variance of estimators in both models.

21.05.2013

Name

Titel

Abstract: A B S T R A C T

28.05.2013

Abstract: tba

04.06.2013

Niklas Willrich (WIAS Berlin)

An extension of finite sample theory for infinite dimensional parameters

Abstract:

11.06.2013

Ekatherina Bezgina (Universität Magdeburg)

Dependence properties of sequential order statistics

Abstract:

18.06.2013

Elmar Diederichs (MIPT Moscow)

Identifying enterotypes by high dimensional clustering

Abstract:

25.06.2013

Hanna Griepentrog (Humboldt-Universität zu Berlin)

Non-commutative Bernstein inequality and its applications to random design regression

Abstract:

02.07.2013

Jim Gatheral (Baruch College, New York)

Market impact with autocorrelated order flow under perfect competition: The Donier model

Abstract:

09.07.13

Prof. Peter Friz (WIAS/TU Berlin)

On the pathwise solution of fully non-linear stochastic partial differential equations.

Abstract: We discuss fully non-linear evolution equations of the type, $du = F(t,x,u,Du,D^u)dt + H(x,Du) \circ dW$ where $W$ is a Brownian sample path (or in fact any continuous path). The geometry associated to the Hamiltonian is seen to play an important role.. Joint work with P. Gassiat (TUB) and T. Souganidis (Chicago).

Datum

Eric Luçon (TU Berlin)

Disordered models of mean-field interacting neurons

Abstract: The dynamics of populations of interacting neurons is often described through models of mean-field interacting diffusions. The diversity of behaviors of such neurons (e.g. inhibitory or excitatory) requires to incorporate in those models a random environment, modeling the heterogeneity of each neuron. The question is then to understand the (quenched or annealed) influence of this random environment on the system in large population (law of large numbers, fluctuations, large deviations), which appears to be nontrivial, even for simple models (Kuramoto model). If time allows, I will also talk about some recent works with W. Stannat on spatially extended versions of these models, which are closer to the modeling of real-world networks of neurons.

23.07.13

Joscha Diehl (TU Diehl)

Rotation invariants of 2D curves and applications to character recognition

Abstract: I introduce a new class of rotation invariants of curves in two dimensional space. These invariants are based on its iterated integrals. The invariants I present are complete in a sense and I describe an algorithm to calculate them. An application to online (stroke-trajectory based) character recognition is given. This seems to be the first time in the literature that the use of the iterated integrals of a curve is proposed for machine learning applications.

Datum

Name

Titel

Abstract: A B S T R A C T

Datum

Name

Titel

Abstract: A B S T R A C T