Date: April 25 - 29, 2022
Venue: HIM lecture hall, Poppelsdorfer Allee 45, Bonn
Organizers: Massimo Fornasier (München), Mauro Maggioni (Baltimore), Holger Rauhut (Aachen)

This meeting is a follow-up workshop to the Trimester Program "Mathematics of Signal Processing".

Signal and data processing as well as machine learning and artificial intelligence (AI) -- all subsumed under the term Data Science -- have sparked technological innovation and are changing our society on many levels. The palette of conceivable applications of these data analysis methods is almost unlimited and we expect that this general-purpose technology will fundamentally change the world we live in. Examples of application areas include (personalized) medicine, optimization of chemical reactions, mechanical engineering, security, agriculture, energy, climate modeling, astro- and particle physics, economics, operations research and portfolio optimization, language processing (speech recognition, translation, speech synthesis), self-driving cars and robotics. Mathematical methods of signal processing are used in many technological devices such as mobile phones, digital cameras, medical, radar and chemical/physical sensors/detectors and more.

The development of mathematical and algorithmic methods for signal processing and machine learning has been the basis for technological breakthroughs of recent years and will continue to drive innovation in the future. The analysis of their potential and reliability is of utmost scientific and societal relevance. From a scientific point of view, signal processing and machine learning have been a constant source for interesting and challenging mathematical problems over the last decades, often leading to new ideas and directions in mathematics. The solution of these problems has provided new and improved approaches and methods to tackle practical signal processing problems, frequently paving the way for significant advances in technology and in the natural sciences.

The workshop will report on recent advances in the mathematics of data science. Particular topics of interest include deep learning, compressive sensing, signal processing and machine learning on graphs as well as applied harmonic analysis.

The lectures given during the workshop will be recorded by default.

Click here for the schedule.

Click here for the abstracts.

Helmut Bölcskei: Lossy Compression on General Data Structure

Karlheinz Gröchenig: Variable Bandwidth and Sampling Theorems

Ron DeVore: Optimal Learning from Data

Felix Voigtländer: Minimax Rates for Learning Classication Functions using Neural Net-
works

Kathlén Kohn: The Geometry of Linear Convolutional Networks

Felix Krahmer: The Convex Geometry of Blind Deconvolution and Matrix Completion Revisited

Nadav Cohen: Generalization in Deep Learning Through the Lens of Implicit Rank Minimization

Soledad Villar: Units-equivariant Machine Learning

Johannes Maly: Covariance estimation from one-bit samples

Alexander Cloninger: Efficient Distribution Classification (Linearized Optimal Transport Embeddings)

Carola Schönlieb: Machine Learned Regularisation for Solving Inverse Problems

Martin Genzel: Solving Inverse Problems With Deep Neural Networks  Robustness Included?

Karin Schnass: Some Simple Inequalities for Rejective Sampling and Why They are Useful

Rayan Saab: New Algorithms for Quantizing Neural Networks

Lorenzo Rosasco: Interpolation and Learning with Scale Dependent Kernels

Richard Kueng: Justifiable quantum advantage for stylized learning challenges