Our research focuses on a wide range of deep learning approaches and their applications to real-world data and issues.

Representation Learning with Deep Neural Networks

Representation Learning Examples

One of our fundamental research direction is representation learning with deep neural networks. Being able to understand, to disentangle, and to control the latent factors of the underlying data would make deep learning more scalable and more trustworthy. It would allow us to gradually move from supervised to unsupervised learning and therefore render model training less dependent on large amounts of labels. Additionally, having control over the latent space of the data provides interpretability of model inference, increase robustness, and allows to characterise model parameter in the weight space.

Analysis and Synthesis of Multimodal Data

Multimodal Data Examples

One of our research goals is to make the large amount of visual and audio content accessible. This includes besides the analysis of such data also the generation of synthetic data with natural characteristics. Currently we focus our research on neural Text-to-Speech (TTS) generation for the German language. In this scope we aim to develop novel approaches for speaker embedding, prosody control, and voice transfer.

Remote Sensing with Deep Neural Networks

Remote Sensing Examples

Remote sensing satellites generate a plethora of data on a daily basis. Given the amount of data and their high complexity, deep learning method have proven extremely useful in their analysis and interpretation. In our lab, we utilize multi-band and multi-modal remote sensing data from a variety of Earth-observing satellites and combine them with complementary data sets to address focused research questions. We apply tailored deep learning models to solve tasks like classification, regression, image segmentation, as well as object detection based on state-of-the-art research results. Our research tackles imminent problems related to climate change, pollution, and disaster relief, as well as socio-economic questions. (more…)

Deep Learning in Assurance & Financial Audit

Financial Data Examples

Our research in this area covers the analysis and forecasting of financial time-series data and audit data analytics. In the field of audit data analytics, we research the interface of deep learning, financial audit and accounting information systems. We aim to develop novel deep learning solutions to enhance conventional audit techniques, based on deep representation learning and privacy-preserving federated learning. Besides, we investigate the malicious misuse of deep learning apporaches to adversarially attack computer assisted audit techniques and misguide auditors. (more…)