This work approaches graph generation (decoding) as the inverse of graph compression (encoding). We show that in a disentanglement-focused deep autoencoding framework, specifically Beta-Variational Autoencoders (Beta-VAE), choices of generative procedures and their parameters arise naturally in the latent space.
Niklas Stoehr 3Emine Yilmaz 3Marc Brockschmidt 6Jan Stuehmer 6
Workshop on Graph Representation Learning, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver
↗ arXiv Paper↗ NeurIPS Paper↗ Presentation Video↗ Interactive Code↗ Poster↗ Slides↗ BibTeX
The proximity between newspapers and political parties is strongly subjective and difficult to measure. Yet, political tendencies of newspapers can have a significant impact on voters’ opinion‐forming and ought to be known by the public in a transparent and timely manner. This article introduces the Sentiment Political Compass (SPC), a data‐driven framework for analyzing political bias of newspapers toward political parties.
Fabian Falck 4Julian Marstaller 7Niklas Stoehr 5Sören Maucher 7Jeana Ren 7Andreas Thalhammer 7Achim Rettinger 7Rudi Studer 7
Policy & Internet, Wiley Online Library, 2019
↗ Pre-Print Paper↗ Journal Paper↗ Code↗ Data↗ Interactive Website↗ Speakerpolitics↗ BibTeX
We present a novel middleware technique called Heatflip, which issues diametrically opposed samples into the temporal and spatial dimensions of the data stored in an external database. Spatial samples provide insights into the temporal distribution and vice versa.
Niklas Stoehr 1Johannes Meyer 1Volker Markl 1Qiushi Bai 2Taewoo Kim 2De-Yu Chen 2Chen Li 2
2018 IEEE International Conference on Big Data (Big Data)
↗ Open Paper↗ IEEE Paper↗ BibTeX
In this work, we thoroughly analyse the ability of text classification models to adapt to transfer learning tasks, whether they are specifically designed for it or not. We directly compare the Transformer Model with an attention-based bi-directional LSTM and naive Logistic Regression as a baseline.
Niklas Stoehr 3Luis Sanchez 3Daniil Gannota 3Fabian Falck 8
2019 IEEE Social Network Management and Security (SNAMS), Second Workshop on Deep and Transfer Learning
↗ IEEE Paper↗ Code↗ Presentation Video↗ BibTeX
In this work, we extend Siamese neural networks to few-shot and zero-shot learning. For few-shot learning, we augment the conventional dual encoder architecture to contain (k+1) encoder branches. For zero-shot learning, we propose Semantic Asymmetric Siamese Networks (SemASN), an asymmetric architecture with two distinct, non-weight sharing encoders on different data modalities (here images and semantic descriptions).
Niklas Stoehr 5 Fabian Falck 5Jiaoyan Chen 5Bernardo Cuenca Grau 5
work in progress
1 TU Berlin 2 University of California, Irvine 3 University College London 4 Imperial College London 5 University of Oxford 6 Microsoft Research 7 KIT 8 Carnegie Mellon University 9 IBM 10 Hertie School of Governance