Training Data Driven Experts in Optimization

TraDE-OPT Logo

Project partner: Silvia Villa (Uni Genova), Lorenzo Rosasco (Uni Genova), Jean-Christophe Pesquet (CentraleSupélec, Paris), Émilie Chouzenoux (CentraleSupélec, Paris), Kristian Bredies (U-GRAZ), Ion Necoara (UPB, Bucharest), Ewa Bednarczuk (PAS, Warsaw), Daniel Węsierski (PAS, Warsaw), Anna Jezierska (PAS, Warsaw), François Glineur (UCL, Louvain), Curzio Basso (CAMELOT)
Project staff: Emanuele Naldi, Lionel Ngoupeyou Tondji

Project funding: EU, A MARIE SKŁODOWSKA-CURIE ACTIONS (MSCA) Innovative Training Network (ITN)
Period: 06/2020 - 05/2024

The main scientific objective of the TraDe-OPT is to derive and analyse efficient optimization algorithms for solving data-driven problems. Applications to a broad range of social, economic, health and urban problems are expected. Nowadays, data production explodes: data are produced by a variety of sensors in industry, vehicles, scanners, internet and mobile devices. One of the emerging challenges is to extract interpretable information from these data. Currently, optimization, and especially, convex optimization, is at the core of many theoretical and algorithmic methods underpinning solver technologies for a myriad of data driven problems.

Project website

Mathematics for machine learning methods for graph-based data with integrated domain knowledge


Project partner: Jochen Garcke (U Bonn, Fraunhofer SCAI), Jan Hamaekers (Fraunhofer SCAI), Gitta Kutyniok (LMU München)
Project staff: Niklas Breustedt (TU Braunschweig)
Project funding: BMBF, Mathematics for innovations
Period: 04/2020 - 03/2023

The goal of this project is, to analyze and develop deep neural networks further to solve real world industry problems such that they allow to incorporate domain knowledge in the architechture of the networks. Such a hybrid approach allow to take advantage of the complementary strength of „end-to-end“ learning and „a priori models/rules“. In this way we aim to achieve much more efficient solutions in many domains of application, for example, by using less data and by achieving consistent model predictions.

Model-based damage analysis

Schadensdetektion durch geführte Ultraschallwellen

Project partner: Natalie Rauter, Rolf Lammering, Wolfgang Weber, Clara Mangalath, Andrey Mikhaylenko (HSU Hamburg)
Project staff: Nanda Kishore Bellam Muradlidhar (TU Braunschweig)

Project funding: DFG, FOR 3022
Period: 08/2020 - 07/2023

Subproject 3 within the research unit 3022, Ultrasonic Monitoring of Fibre Metal Laminates
Using Integrated Sensors

This research unit investigates how to do automated health monitoring off fibre metal laminates with guided ultrasonic waves. This subproject deals with the numerical modelling of the wave propagation in fibre metal laminates and the inverse problem of damage reconstruction based on a few vibration measurements.

Project website

Bilevel Optimal Transport

Optimal transport plan

Project partner:
Prof. Christian Meyer, Sebastian Hillbrecht (Fakultät für Mathematik, TU Dortmund), Hinrich Mahler (Institut für Analysis und Algebra, TU Braunschweig)
Project funding: DFG, SPP1962
Period: 10/2019 - 09/2022
In this project we analyze bilevel optimization problems in which the lower-level problem is a problem of optimal transport. We treat both the Kantorovich form and the Beckmann form of the problems, since these problem allow to optimze for different quantities. In the first case we can optimize, for example, the transport cost for given starting and endpoint while in the second case we can optimiza the transport cost at some given point in space.

Dequantization of speech signals

Project partner:
Prof. Timo Gerkmann (U Hamburg), Prof. Tim Fingscheidt (TU Braunschweig), Dr. Ziyue Zhao (TU Braunschweig), Dr. Christoph Brauer (TU Braunschweig)
Period: 01/2016 - 04/2019
In this project we use and develop mathematical models to enhance speech signals which have been quantized coarsly for, for example, wireless transmission. This happens, for instance, in wireless telephones and with hearing aids. The aim is, to enhance the speech quality by using methods from compressed sensing and machine learning.

SPEAR - Sparse Exact and Approximate Recovery

Project partner:

Prof. Marc Pfetsch, Andreas Tillmann, Institute for Mathematical Optimization, TU Braunschweig, Christian Kruschel, Institut for Analysis and Algebra, TU Braunschweig
Project funding: DFG
Period: 04/2011 - 04/2013
This project is concerned with the problem of finding sparse solutions to linear systems of eqautions. It is located at the border between analysis and combinatorial optimization. The main goal of our project is to obtain a better understanding of the conditions under which (efficiently) finding such a sparse solution, i.e., recovery, is possible.


Sparsity and Compressed Sensing in Inverse Problems

Project partner:
Prof. Gerd Teschke, Dr. Evelyn Herrholz, Hochschule Neubrandenburg
Project funding: DFG, SPP 1324
Period: 06/2008 - 06/2011
This project aims at a thorough theory of compressed sensing for ill-posed problems and is hence located at the intersection of signal processing and ill-posed problems. Compressed sensing is a promising new field which tries to tackle the problem of high-dimensional data by combining the measuring and the compression step into one single process of "compressive sampling". The theory is well developed for well-posed finite dimensional linear problems. The main points to be addressed in this project are a proper formulation in infinite dimensional spaces and the treatment of ill-posed operators (e.g.~compact operators) - both linear and non-linear.

Image sequence interpolation using optimal control

Project partner:
Kanglin Chen, Zentrum für Technomathenatik, Uni Bremen
Project funding: ZF Uni Bremen, Doktorandengruppe SCiE
Period: 11/2008 - 11/2011
In this project we develop method to generate an "interpolating movie" between two given images. We model this problem as a problem of optimal control with a transport equation as a constraint. We especially interested to transport images with discontinuities (edges) and also aim at low regularity of the flow field.