Fachbereich Mathematik

Kolloquien WS 2020/21


Vorheriges Semester Nächstes Semester

Montag, 07.12.2020: Arrow of Time in Physics

Prof. Dr. Peter Pickl (LMU München)

Since the work of Boltzmann, there is the important question of how to explain the arrow of time in statistical physics: While most microscopic laws do not know a direction of time, one can clearly see an arrow of time in macroscopic nature. A glass that falls on the floor breaks into many pieces, but the same process is not observed backwards. The usual attempts to explain this are based on a very special initial condition, a state of very great order. This state then naturally develops into a state of great disorder. The asymmetry (order -> disorder) over time is not surprising. However, this explanation shifts the problem to the question of where the initial special condition came from. Alternative explanations are to be discussed in the lecture. Indeed, considering systems without equilibrium, it seems possible to find models for which the entropy typically increases. The focus in the talk is on the concepts, many of the questions asked are still "work in progress" from a mathematically rigorous point of view.

Uhrzeit: 16:00
Ort: online - wenn Sie Zugang haben wollen, schicken Sie bitte eine Nachricht an Angelika Spörer-Schmidle
Gruppe: Kolloquium
Einladender: Cederbaum

Montag, 07.12.2020: Vortrag in der Reihe "Mathematiker im Beruf"

Christian Laus (Braincourt)

In der Vortragsreihe "Mathematiker im Beruf", die sich vor allem an die Studierenden des Fachbereichs Mathematik richtet, berichten ehemalige Studierende des Fachbereichs über ihren Werdegang, ihr jetziges Arbeitsfeld und wie ihnen ihr Mathematikstudium dabei zu gute kommt. -- Künstliche Intelligenz überflügelt den Menschen, selbstlernende Maschinen wissen mehr über uns als unsere Freunde und Daten sind das neue Öl - Die Themenfelder aus dem Bereich Data Science sind in der alltäglichen Berichtserstattung nicht mehr wegzudenken. -- Christian Laus ist Leiter des Data Science & Big Data Teams des Beratungshauses Braincourt und begleitet seit seinem Studium verschiedene Firmen vom Mittelstand bis Dax-Unternehmen auf dem digitalen Weg. -- Im Rahmen der Vortragsreihe "Mathematiker:innen im Beruf" beantwortet er die Fragen, was ein Data Scientist tatsächlich tut, welche Skills aus der Mathematik braucht man und warum es der "sexiest job oft the 21st century" ist. -- Anschließend an den Vortrag und der Fragerunde gibt es mithilfe von Open Source Technologien praktische Einblicke in den Entstehungsprozess einer Deep Learning KI zur Bilderkennung. Christian Laus präsentiert ein "Convolutional Neural Network" zur Bilderkennung, welches beispielsweise verschiedene mathematische Formeln unterscheidet. Dabei erklärt er die wichtigsten Tricks dieser KI und berichtet von allgemeinen Erfahrungen zu dem Thema Deep Learning in der Wirtschaft.

Uhrzeit: 18:30 - 19:15
Ort: Online-Format: https://zoom.us/j/97101758202
Gruppe: Kolloquium
Einladender: Fachschaft Mathematik + Studiendekan

Montag, 14.12.2020: Vortrag in der Reihe "Mathematiker im Beruf" - Mathematik und Softwareentwicklung: Yin und Yang oder Öl und Wasser?

Dr. Michael Sperber (Geschäftsführer der Active Group GmbH)

In der Vortragsreihe "Mathematiker im Beruf", die sich vor allem an die Studierenden des Fachbereichs Mathematik richtet, berichten ehemalige Studierende des Fachbereichs über ihren Werdegang, ihr jetziges Arbeitsfeld und wie ihnen ihr Mathematikstudium dabei zu gute kommt. - Zum Unternehmen: Die Active Group GmbH ist eine Softwarefirma in Tübingen, die hauptsächlich Individualsoftware entwickelt, und zwar in allen möglichen Branchen: Sozialpädagogik, Halbleiterfabrikation, Bau, Versicherung, Banken, und noch einiges dazwischen. Sie ist führend in der Anwendung funktionaler Programmierung, einer Sammlung von besonders effektiven Techniken und Technologien in der Softwareentwicklung. - Zum Vortrag: "Viele unserer Mitarbeiter:innen haben eine Informatik-Ausbildung, aber ein paar Mathematik-Abschlüsse sind auch dabei. Ein Hintergrund in Mathematik ist natürlich besonders bei analytischen Aufgabenstellungen relevant, aber Mathematik ist auch in den alltäglichen Aspekten der Softwareentwicklung wichtig: Wir wenden routinemäßig mathematische Techniken in der Modellierung an, aber auch wenn es darum geht, die Zuverlässigkeit von Software sicherzustellen. Besonders relevant sind dafür Algebra und zunehmend auch Kategorietheorie. Trotzdem sind Mathematiker:innen nicht automatisch Softwareentwickler:innen. Es gibt wichtige kulturelle und pragmatische Unterschiede. Insbesondere ist Mathematik in einigen Software-Kreisen regelrecht verpönt. Der Vortrag beleuchtet, worauf es ankommt, wenn Mathematiker:innen Software entwickeln."

Uhrzeit: 18:30 - 19:15
Ort: Online-Format: https://zoom.us/j/91495123170
Gruppe: Kolloquium
Einladender: Fachschaft Mathematik + Studiendekan

Montag, 11.01.2021: Curve counting and tropical geometry

Prof. Dr. Hannah Markwig (Universität Tübingen)

The talk introduces tropical geometry as a tool for enumerative geometry. In enumerative geometry, we study counts of curves satisfying certain conditions. Particularly, we consider generating series for important enumerative invariants. Some of them appear in the context of mirror symmetry and can be related to Feynman integrals. This can be used to study their properties such as quasimodularity.

Uhrzeit: 17:15
Ort: online - wenn Sie Zugang haben wollen, schicken Sie bitte eine Nachricht an Angelika Spörer-Schmidle
Gruppe: Kolloquium
Einladender: Cederbaum

Dienstag, 12.01.2021: Talk in the series "Mathematicians in industry and buisness"

PD Dr. Birger Horstmann (Deutsches Zentrum für Luft- und Raumfahrt)

In the lecture series "Mathematicians in industry and buisness", which is aimed primarily at students of the mathematics department, former students (of the department) report on their careers, their current field of work and how their mathematics studies will benefit them. - DLR is the research center for aerospace of the Federal Republic of Germany. Some 8,000 employees work together on a unique variety of topics in aviation, aerospace, energy, transportation, digitization and security. DLR's Institute of Technical Thermodynamics conducts research in the field of efficient and resource-saving next-generation energy storage systems. The Department of Computational Electrochemistry at the Helmholtz Institute Ulm for Electrochemical Energy Storage (HIU) develops new theories and simulation tools for battery materials and cells. - About the lecture: Theoretical physicists, (physical) mathematicians and engineers work together at DLR to improve batteries. First, models for the ion transport in electrolytes and the movement of phase boundaries are created on the basis of statistical physics and fluid dynamics. These systems of partial differential equations are then programmed in Matlab, Python or C / C ++ and the physico-chemical processes are simulated - from the nanoscale to the macroscale. In the lecture, the speaker will present all work steps from the modeling idea through numerical implementation to simulation evaluation.

Uhrzeit: 18:00 - 18:45
Ort: Online-Format: https://bbb03.zdv.uni-tuebingen.de/b/cym-ck7-rup-c77
Gruppe: Kolloquium
Einladender: Fachschaft Mathematik + Studiendekan

Montag, 08.02.2021: First-order Optimization Algorithms for Machine Learning

Prof. Dr. Peter Ochs (Universität Tübingen)

Optimization problems in the context of Machine Learning are usually high-dimensional and are naturally modelled as non-smooth optimization problem. This setting requires the development of dedicated optimization algorithms and makes the usage of classical black-box algorithms impractical. Nevertheless, the special structure of the optimization problems allows for efficient optimization with first-order algorithms at the forefront. Such algorithms rely on first order (sub-)derivative information about the objective function with a suitable concept that generalizes differentiation to non-smooth functions. In this talk, we introduce the common structure that is exploited by first-order algorithms for non-smooth optimization problems. We discuss the convergence of so-called Proximal Gradient Algorithms including an analysis of their convergence rates. Moreover, we develop an approach towards generalizing quasi-Newton methods to the regime of non-smooth optimization.

Uhrzeit: 17:15
Ort: online - wenn Sie Zugang haben wollen, schicken Sie bitte eine Nachricht an Angelika Spörer-Schmidle
Gruppe: Kolloquium
Einladender: Cederbaum, Huisken, Kröncke