Anthony Richardson, M. Sc.

Master Thesis: Enhanced Video Question-Answering on Unpaired Video Data

The goal of this Master Thesis is to create a machine learning model for Video Question-Answering, that can leverage large amounts of unpaired video data through unsupervised learning.

More information can be found here.

Master Thesis: Online Hyperparameter Adaptation via Learned Controllers

The goal of this Master Thesis is to pair a machine learning  model with a light-weight controller that dynamically adjusts the used training hyperparameters throughout the course of the training.

More information can be found here. 

CogniFuse and Multimodal Deformers: A Unified Approach for Benchmarking and Modeling Biosignal Fusion

Full paper: [pdf]

Authors: Anthony Richardson, Michael Beetz, Tanja Schultz, Felix Putze

Conference: 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2025)

Abstract: Continuously monitored physiological signals carry rich information about the human body and the biological processes happening within. Extracting this information from casually collected biosignal data in activities of daily living holds great potential for real-time monitoring of physical and mental states, but comes with increased difficulty due to the influence of noise and artifacts. Thus, we create CogniFuse, the first publicly available multi-task benchmark for multimodal biosignal fusion in such challenging environments. For many biosignals, especially electrophysiological signals, the information contained in different frequency bands plays a significant role in analyzing the physiological states of the body. Therefore, we introduce a group of novel fusion models, called Multimodal Deformers, that capture multi-level power features as well as long- and short-term temporal dependencies in multimodal biosignal data. In particular, our proposed Multi-Channel Deformer achieves the highest average benchmark score, outperforming all models of comparison. To assure full transparency and reproducibility, and to support future research on multimodal biosignal fusion, all code and data is made publicly available.

Motion Diffusion Autoencoders: Enabling Attribute Manipulation in Human Motion Demonstrated on Karate Techniques

Full paper: [pdf]

Authors: Anthony Richardson, Felix Putze

Conference: 27th International Conference on Multimodal Interaction (ICMI '25)

Abstract: Attribute manipulation deals with the problem of changing individual attributes of a data point or a time series, while leaving all other aspects unaffected. This work focuses on the domain of human motion, more precisely karate movement patterns. To the best of our knowledge, it presents the first success at manipulating attributes of human motion data. One of the key requirements for achieving attribute manipulation on human motion is a suitable pose representation. Therefore, we design a novel continuous, rotation-based pose representation that enables the disentanglement of the human skeleton and the motion trajectory, while still allowing an accurate reconstruction of the original anatomy. The core idea of the manipulation approach is to use a transformer encoder for discovering high-level semantics, and a diffusion probabilistic model for modeling the remaining stochastic variations. We show that the embedding space obtained from the transformer encoder is semantically meaningful and linear. This enables the manipulation of high-level attributes, by discovering their linear direction of change in the semantic embedding space and moving the embedding along said direction. All code and data is made publicly available.