Forschungskolloquium Wirtschaftsinformatik

Forschungskolloquium Wirtschaftsinformatik

16. Dez.

Im Rah­men des For­schungs­kol­lo­qui­ums Wirt­schafts­in­for­ma­tik und Data Sci­ence re­fe­rieren am 16. Dezember 2021 ab 16:15 Uhr, Yannick Rudolph (Doktorand in der Arbeitsgruppe Maschinelles Lernen von Prof. Dr. Ulf Brefeld an der Leuphana Universität Lüneburg und Mitarbeiter der SAP SE, Berlin) und Jonas Scharfenbverger (Wiss. Mitarbeiter an der Leuphana Universität Lüneburg, Professur für Wirtschaftsinformatik).

Yannick Rudolph

“Modeling Conditional Dependencies in Multiagent Trajectories
We study modeling joint densities over sets of random variables (next-step movements of multiple agents) which are conditioned on aligned observations (past trajectories). For this setting, we propose an autoregressive approach to model intra-timestep dependencies, where distributions over joint movements are represented by autoregressive factorizations. In our approach, factors are randomly ordered and estimated with a graph neural network to account for permutation equivariance, while a recurrent neural network encodes past trajectories. We further propose a conditional two-stream attention mechanism, to allow for efficient training of random factorizations. We experiment on trajectory data from professional soccer matches and find that we model low frequency trajectories better than variational approaches.

Jonas Scharfenberger

Like other disciplines, Information Systems is experiencing a growing volume of scholarly publications. This development exacerbates the threat of conceptual fragmentation. Previously, solutions based on repositories and databases were suggested to combat this issue, but the effort needed to build and maintain these solutions has impeded their widespread adoption. In response, the literature is exploring machine- learning-based approaches. We join this exploration proposing a computer vision approach to detecting conceptual models and extracting their constituents. The developed tool can serve as a foundation for automating the population of scientific databases describing theoretical models. We evaluate our deep learning approach against a sample of papers containing graphical theoretical models, and show that 81.5% of all constructs, items, and path coefficients can be correctly classified. This has the potential to significantly reduce manual efforts to populate scientific databases and can be an important step towards the augmentation of the work of theorists.

Zugangsdaten:

https://leuphana.zoom.us/j/99663552105?pwd=dG9KZUtJU0hUMkE4RHNEVDhiUlNIZz09

Meeting-ID: 99663552105
Passwort: FoKoWI