Seminarier

Se alla kommande seminarier i LäsIT samt seminariesidor hos doktorandseminariets hemsida, TDB, Vi2, Theory and Applications Seminars (TAS) @ UpMARC, Matematiska institutionen och The Stockholm Logic Seminar.

Seminar
Måndag 27 maj
Arno Solin, Aalto University: Know Your Boundaries: Constraining Gaussian Processes by Variational Harmonic Features
Plats: ITC 2244, Tid: 11:15

Abstract: Gaussian processes (GPs) provide a powerful framework for extrapolation, interpolation, and noise removal in regression and classification. This paper considers constraining GPs to arbitrarily-shaped domains with boundary conditions. We solve a Fourier-like generalised harmonic feature representation of the GP prior in the domain of interest, which both constrains the GP and attains a low-rank representation that is used for speeding up inference. The method scales as O(nm^2) in prediction and O(m^3) in hyperparameter learning for regression, where n is the number of data points and m the number of features. Furthermore, we make use of the variational approach to allow the method to deal with non-Gaussian likelihoods. The experiments cover both simulated and empirical data in which the boundary conditions allow for inclusion of additional physical information.
This is joint work with Manon Kok (TU Delft) and it was recently presented at AISTATS 2019.

Seminar
29 maj
Simon Olofsson, Imperial College London: Design of experiments for model discrimination using Gaussian process surrogates
Plats: ITC 2345, Tid: 13:15

Many researchers and engineers study noisy processes with mechanisms and reactions that cannot be observed directly, e.g. in pharmaceutical development or chemical products manufacturing. Researchers hypothesise rival mechanistic models to explain observations, but may have insufficient data to discriminate between the hypotheses, i.e. discard inaccurate models. Optimal experimental design suggests experimental conditions maximising the expected utility from additional experiments to aid mechanistic model discrimination.

Traditional approaches for design of experiments rely on (i) Taylor expansions of the mechanistic models, or (ii) Monte Carlo methods. However, for mechanistic models that are expensive-to-evaluate black boxes, both approaches may be computationally prohibitive. We propose a surrogate model approach, where we place Gaussian process priors on the black-box models. We show that the surrogate model approach allows us to design optimal static and dynamic experiments for discriminating black-box mechanistic models.

This is joint work with Marc P. Deisenroth and Ruth Misener.

Disputation | Dissertation
4 juni
Tomas Wilkinson: Learning based Word Search and Visualisation for Historical Manuscript Images
Plats: TLS (Tidskriftläsesalen), Carolina Rediviva, Tid: 10:15

Tomas Wilkinson will present and defend his PhD thesis Learning based Word Search and Visualisation for Historical Manuscript Images.
Opponent: Josep Lladós, Universitat Autònoma de Barcelona, Computer Vision Center.
Supervisor: Anders Brun.

DiVA includes an abstract and the full text of the thesis.

Licentiatseminarium | Licentia
5 juni
Anton Artemov: Inverse Factorization in Electronic Structure Theory: Analysis and Parallelization
Plats: ITC 2347, Tid: 10:15

External examiner is Associate Professor Lars Karlsson from Umeå University

Abstract: This licentiate thesis is a part of an effort to run large electronic structure calculations in modern computational environments with distributed memory. The ultimate goal is to model materials consisting of millions of atoms at the level of quantum mechanics. In particular, the thesis focuses on different aspects of a computational problem of inverse factorization of Hermitian positive definite matrices. The considered aspects are numerical properties of the algorithms and parallelization. Not only is an efficient and scalable computation of inverse factors necessary in order to be able to run large scale electronic computations based on the Hartree-Fock or Kohn-Sham approaches with the self-consistent field procedure, but it can be applied more generally for preconditioner construction. Parallelization of algorithms with unknown load and data distributions requires a paradigm shift in programming. In this thesis we also discuss a few parallel programming models with focus on task-based models, and, more specifically, the Chunks and Tasks model.

Disputation | Dissertation
5 juni
Amit Suveer: Methods for Processing and Analysis of Biomedical TEM Images
Plats: ITC 2446, Tid: 10:15

Amit Suveer will present and defend his PhD thesisMethods for Processing and Analysis of Biomedical TEM Images.
Opponent: Professor Jari Hyttinen, Tampere University of Technology, Faculty of Biomedical Sciences and Engineering.
Supervisors: Ida-Maria Sintorn, Carolina Wählby, Natasa Sladoje.

DiVA includes an abstract and the full text of the thesis.

Se också listan över alla kommande seminarier.

Internt arrangerade seminarier. Föreläsare kan vara både interna och externa.

RSS