Data Science Talk Series Summer Semester 2021

 
We are happy to announce that our Talk Series will be continued online via Zoom. Today we would like to invite you to our next sessions:

● Friday 5 March 2021 @ 14:00

Sylvia Kritzinger (Department of Government)

Exploring possibilities for dealing with survey data: data science and political science in partnership?

● Friday 26 March 2021 @ 12:00

Nils Kriege (Research Group Data Mining and Machine Learning)

Learning with Structured Data by Matching and Embedding Graphs

● Friday 16 April 2021 @ 12:00

O. Anatole von Lilienfeld (Computational Materials Physics)

Quantum Machine Learning

Join via Zoom

https://zoom.us/j/95484956002?pwd=ZXcreERLZkk3ejdRN1hvRzcrTGY0QT09

Meeting-ID: 954 8495 6002

Password: dsunivie21

Friday 5 March 2021 @ 14:00

Sylvia Kritzinger (Department of Government)

Exploring possibilities for dealing with survey data: data science and political science in partnership?

Political science research on public opinion and voting behaviour focuses on the emergence and development of political attitudes and political behaviours. We thereby analyse both long-term factors such as individual (psychological) predispositions and short-term implications stemming from the private environment of voters, the media landscape as well as the party communication per se. To conduct this analysis, data is collected via representative citizen surveys. Both the data collection and the subsequent statistical modelling of the data are strongly theory-driven. For our research, we can in the meantime rely on a large number of (panel) data sets, where a large number of variables per individual is available. However, for each of our analyses we normally use very few of those variables. In the presentation, I will first outline our research interests by presenting some examples. Second, I will present the various data sets and their characteristics (including linkage options). By doing so, overlaps with research interests of the data sciences should (hopefully) materialize.

Friday 26 March 2021 @ 12:00

Nils Kriege (Research Group Data Mining and Machine Learning)

Learning with Structured Data by Matching and Embedding Graphs

Data mining and machine learning for structured data is becoming increasingly important in domains such as social network analysis, computer vision or chem- and bioinformatics. In this talk, I give an overview of my work in this area with a focus on applications in cheminformatics. The talk is divided into three closely connected parts.
The maximum common subgraph problem asks for a largest substructure that is contained in two given graphs. The problem is NP-hard in general. I introduce polynomial-time algorithms for trees and tree-like graphs. Motivated by constraints relevant in cheminformatics a variation of the problem is formalized and solved efficiently in series-parallel graphs.
Graph kernels are specific similarity measures for graphs, which enable the application of established machine learning approaches such as support vector machines to graphs. I will present kernels based on Weisfeiler-Lehman refinement.
Graph neural networks extend deep learning techniques, which have been proven to be extremely successful for data such as images, to directly operate on graphs. I will present a technique for deep graph matching, which learns and refines feature representations to reach a consensus mapping.

Friday 16 April 2021 @ 12:00
O. Anatole von Lilienfeld (Computational Materials Physics)

Quantum Machine Learning

Many of the most relevant observables of matter depend explicitly on atomistic and electronic details, rendering a first principles approach to computational materials design mandatory. Alas, even when using high-performance computers, brute force high-throughput screening of material candidates is beyond any capacity for all but the simplest systems and properties due to the combinatorial nature of compound space, i.e. all the possible combinations of compositional and structural degrees of freedom. Consequently, efficient exploration algorithms exploit implicit redundancies and correlations. I will discuss recently developed statistical learning based approaches for interpolating quantum mechanical observables throughout compound space. Numerical results indicate promising performance in terms of efficiency, accuracy, scalability and transferability.