Heidelberg Graduate School HGS MathComp

HGS MathComp Events

[SS 15] [WS 15] [SS 16] [WS 16] [SS 17] [WS 17] [SS 18] [WS 18] [SS 19]

Sommersemester 2019
Compact Course
info  Spatial and Temporal Analyses of Geographic Phenomena (STAP19)Various SpeakersApril 1-4, 2019ECTS-Points: 2
Abstract, registration & information:
Katharina Anders, Bernhard Höfle, Hubert Mara

Compact Course & Workshop:
- Automatic methods for 3D geospatial data processing
- Geographic applications of 3D data analysis
- Hands-on: 3D point cloud and mesh analysis
- Programming and research challenge: Development of computational methods for 3D information extraction

Invited speakers:
- Prof. Dr. Andreas Nüchter, University of Würzburg
- Jorge Martínez Sánchez, University of Santiago de Compostela

Please register on the website of the Compact Course until February 15, 2019

Project Auto3Dscapes:

Contact: Katharina Anders


Link for more information
Mathematikon, Conference Room, 5th Floor, Room 5/104, Im Neuenheimer Feld 205, 69120 Heidelberg
IWR Colloquium
info  The Statistical Finite Element MethodProf. Mark A. GirolamiJuly 3, 2019 / 16:15ECTS-Points: not yet determined

Abstract (PDF)

Link for more information
Mathematikon, Conference Room, Room 5/104, 5th Floor, Im Neuenheimer Feld 205, 69120 Heidelberg
not yet determined
info  IWR School "A Crash Course in Machine Learning with Applications in Natural- and Life Sciences (ML4Nature)"Various SpeakersSeptember 23-27, 2019ECTS-Points: not yet determined
Abstract, registration & information:
The IWR School 2019 gives a crash course in machine learning with applications from Natural Sciences and Life Sciences. We target young researchers from Natural Sciences and Life Sciences who want to learn more about machine learning. A background in machine learning is not required. Besides introducing the basic concepts of machine learning, we teach selected topics in more depth, such as deep learning, metric learning, transfer learning, Bayesian inverse problems, and causality. Experts from machine learning, Natural Science and Life Science explain how these machine learning approaches are utilized to solve problems in their respective fields of research.

Target Audience:

Postgraduate students, PhD candidates, postdocs and young researchers:

- from Natural and Life Sciences: Microscopy, Biology, Medical, Physics,…
- with interest in Machine Learning
- Master students from Heidelberg University (core course listed in LSF)


The IWR School 2019 is taught in a series of courses and single lectures by:

- Christoph Lampert, Institute of Science and Technology Austria
- Oliver Stegle, European Bioinformatics Institute
- Robert Scheichl, Heidelberg University
- Dominik Janzing, Max Planck Institute for Intelligent Systems
- Klaus Maier Hein, German Cancer Research Center
- Bjoern Ommer, Heidelberg University
- Ullrich Köthe, Heidelberg University
- Anna Kreshuk, European Molecular Biology Laboratory

For more information please visit the website of the IWR School 2019.

Link for more information
Mathematikon, Conference Room, 5th Floor, Room 5/104, Im Neuenheimer Feld 205, 69120 Heidelberg
not yet determined
info  A Sequential Homotopy Method for Mathematical Programming ProblemsDr. Andreas PotschkaMay 16, 2019 / 14:15ECTS-Points: not yet determined
Abstract, registration & information:
We consider nonconvex and highly nonlinear mathematical programming problems including finite dimensional nonlinear programming problems as well as optimization problems with partial differential equations and control constraints. We present a novel numerical solution method, which is based on a projected gradient/anti-gradient flow for an augmented Lagrangian on the primal/dual variables. We show that under reasonable assumptions, the nonsmooth flow equations possess uniquely determined global solutions, whose limit points (provided that they exist) are critical, i.e., they satisfy a first-order necessary optimality condition. Under additional mild conditions, a critical point cannot be asymptotically stable if it has an emanating feasible curve along which the objective function decreases. This implies that small perturbations will make the flow escape critical points that are maxima or saddle points. If we apply a projected backward Euler method to the flow, we obtain a semismooth algebraic equation, whose solution can be traced for growing step sizes, e.g., by a continuation method with a local (inexact) semismooth Newton method as a corrector, until a singularity is encountered and the homotopy cannot be extended further. Moreover, the projected backward Euler equations admit an interpretation as necessary optimality conditions of a proximal-type regularization of the original problem. The prox-problems have favorable properties, which guarantee that the prox-problems have uniquely determined primal/dual solutions if the Euler step size is sufficiently small and the augmented Lagrangian parameter is sufficiently large. The prox-problems morph into the original problem when taking the step size to infinity, which allows the following active-set-type sequential homotopy method: From the current iterate, compute a projected backward Euler step by applying either local (inexact) semismooth Newton iterations on the step equations or local (inexact) SQP-type (sequential quadratic programming) methods on the prox-problems. If the homotopy cannot be continued much further, take the current result as a starting point for the next projected backward Euler step. If we can drive the step size all the way to infinity, we can transition to fast local convergence. We can interpret this sequential homotopy method as extensions to several well-known but seemingly unrelated optimization methods: A general globalization method for local inexact semismooth Newton methods and local inexact SQP-type methods, a proximal point algorithm for problems with explicit constraints, and an implicit version of the Arrow--Hurwicz gradient method for convex problems dating back to the 1950s extended to nonconvex problems. We close the talk with numerical results for a class of highly nonlinear and badly conditioned control constrained elliptic optimal control problems with a semismooth Newton approach for the regularized subproblems.

Preprint available on https://arxiv.org/abs/1902.06984

Link for more information
Mathematikon, Room 2/414, 2nd Floor, Im Neuenheimer Feld 205, 69120 Heidelberg
not yet determined