Submitted by jolanta.pielasz... on January 4, 2019 - 16:10
Speaker:
Prof. S. Ejaz Ahmed from Brock University, Canada
We are pleased to announce that the Special seminar on Big Data will be held just after the Multivariate and Mixed Linear Models conference, on Saturday, 4th of May in Bedlewo, Poland.
Submitted by xavierdeluna on January 17, 2019 - 14:26
Speaker:
See below
The Conference starts with a Welcome reception, approximately at 20.00 on Sunday, March 10, and is concluded with lunch at 11.00 on Thursday, March 14.
Approximately 18 hours will be scheduled of which 12 lectures will be given by the main speakers.
Lectures will be given in the morning, in late afternoon and in the evening, with opportunity for outdoor activities during midday.
The Conference Programme will be available in February.
Preliminary topics:
AdaNet: adaptive learning of neural networks (Corinna Cortes, Google Research)
Submitted by umberto.picchini on November 17, 2018 - 09:30
Speaker:
Josef Wilzén (Linköping)
Inference from fMRI data faces the challenge that the hemodynamic system, that relates the underlying neural activity to the observed BOLD fMRI signal, is not known. We propose a new Bayesian model for task fMRI data with the following features: (i) joint estimation of brain activity and the underlying hemodynamics, (ii) the hemodynamics is modeled nonparametrically with a Gaussian process (GP) prior guided by physiological information and (iii) the predicted BOLD is not necessarily generated by a linear time-invariant (LTI) system.
Submitted by umberto.picchini on October 25, 2018 - 16:49
Speaker:
Thomas Schön (Uppsala University)
Abstract: In this talk I will focus on one of our recent developments where we show how the Gaussian process (GP) can be used to solve stochastic optimization problems. Our main motivation for studying these problems is that they arise when we are estimating unknown parameters in nonlinear state space models using sequential Monte Carlo (SMC). The very nature of this problem is such that we can only access the cost function (in this case the likelihood function) and its derivative via noisy observations, since there are no closed-form expressions available.
Submitted by umberto.picchini on October 24, 2018 - 08:32
Speaker:
Harri Lähdesmäki
Welcome to a seminar with Harri Lähdesmäki (Aalto) on 25 October at 13.15 in room MV:L14 at Dept. Mathematical Sciences (Chalmers tvärgata 3, Gothenburg).
Submitted by jolanta.pielasz... on March 12, 2018 - 09:35
Speaker:
Rauf Ahmad, Uppsala University
———— ———— ———— ———— ———— ————
Title: On the risk of LASSO type estimators
———— ———— ———— ———— ———— ————
Abstract:
This talk attempts to address the issues of sparsity and mathematically intractable risk of LASSO estimator - by introducing a parallel LASSO-type estimator which does not rest on sparsity and whose risk is exactly computable. It offers a generalization of the classical LASSO estimator for non-orthogonal design, allowed to be of less than full rank. We compare the risk of new estimator with its shrinkage competitor, the ridge estimator.
Submitted by jolanta.pielasz... on December 5, 2017 - 11:05
Speaker:
Professor Aila Särkkä, Chalmers University of Technology
Abstract: Analysis of deep polar ice cores has become an important tool for deriving climate information from the past. Interpretation of ice core records requires an accurate dating of the ice. The recent dating relies on models where the key element is the simulation of the individual history of ice deformation for each specific core site. We present a two-stage method for the estimation of the deformation history in polar ice using the measured anisotropy of air inclusions from deep ice cores.
Submitted by jolanta.pielasz... on September 19, 2017 - 12:35
Speaker:
Niharika Gauraha
Abstract: We discuss an important research problem concerning selection of covariates that are both highly informative and correlated with each other. Most of the variable selection methods exclude the highly correlated covariates from the regression equations, which can lead investigators to overlook significant features. Especially in high dimensional settings, independence assumptions among predictors are unlikely to be fulfilled in general.