Student Thesis Presentations October 2022

Date: Oct 26, 2022, 4:15 pm

Location: online



4:15 - 4:30


  Jonas Freitag


  Unsupervised Representation Learning for Automatic Analysis of Sperm Cell Motility



4:30 - 4:45


  Dominik Süsser


  Substitution-based Emotional Style Transfer in Tweets Using Transformers

Automatic Analyses of Multi-Agent Interactions towards Behaviour of True Artificial Peers

Lecturer: Dr.-Ing. Dipl.-Inf. Ronald Böck

Date: Oct 20, 2022, 2:30 pm

Location: Building F1, Room 004



Assessments and investigations in multi-agent or multi-party interactions, which are interactions of multiple human users and multiple technical systems that form groups of equal partners, are emerging but challenging topics in computer sciences.  This is particularly the case when systems are integrated in the daily lives of users – subsuming private as well as public/workplace environments. For this, a magnitude of signals has to be detected, distinguished, classified, and interpreted on various time scales for a proper understanding of the current situation, especially 1) activities and (re-)actions as well as 2) social and affective signals of the communication partners. In group settings, this includes additionally the handled tasks as well as the entire group’s characteristics. Furthermore, appropriate reactions of the technical systems and automatic devices, being equal partners in the communication, are necessary and important to establish a valuable collaborative and cooperative interaction.


Regarding group assessment, we consider three aspects:

1) At first, we are interested in multi-modal investigations of interaction patterns in multi-agent interactions, which are briefly discussed in this talk. For this, we analyse the “CASIA Coffee House” Corpus, comprising naturalistic interactions of a human user and two artificial agents. Given the material, i) prototypical movement patterns in multi-agent interactions could be derived and ii) typical interaction behaviour, known from social sciences, were identified and automatically detected.

2) Further, we focus on analyses of groups, mainly investigating the performance based on acoustic cues in group meetings, being linked to social and affective signals. For this, the “Parking Lot Corpus” is analysed, which provides group discussions on the improvement of the current parking situation on a Mid-Western US university and reflects a multi-human communication. We show that prosodic and acoustic features provide significant discriminative power for the assessment of performing groups, especially regarding multiple performance measures, usually applied in meeting sciences. Furthermore, we focus on analyses of the meeting effectiveness since this measure can be used to evaluate both, the “objective” outcome of the meeting as well as the perceived (“subjective”) outcome of the group in the sense of an overall satisfaction. Automatic investigations allow operationalised assessments of the entire group or particular group members and provides options for an “online” supervision during the group meeting.

3) Finally, the way of appropriate generation of actions and reactions is discussed. If the technical system is not being intended to act as a simple assistant, a rather flexible approach is necessary being beyond a simple rule-based action creation. For this, we highlight an approach how technical systems can be equipped with an own behaviour, comprising very own objectives and characteristics. Therefore, we are on the way towards technical systems and devices being considered as True Artificial Peers.

Additionally, the talk presents an overview on the current projects of the group, being related to natural language processing and sentiment analysis.


Bio: Ronald Böck is currently Head of Research at Genie Enterprises Inc, Branch Office Germany. Before, Dr Böck was Privatdozent (post-doctoral researcher) in the Cognitive Systems Group, Institute of Information and Communication Technology, at the Otto von Guericke University Magdeburg, Germany. He is Computer Scientist (Technical University Ilmenau, Germany) and holds the Doktoringenieur (PhD) and the Habilitation (Higher Doctorate) from the Otto von Guericke University Magdeburg, Germany. Ronald Böck is member of various multi-disciplinary scientific initiatives and collaboration project between industrial and academic partners. In 2012, he stayed as guest scientist at the Trinity College Dublin, Ireland. In 2015, he was Visiting Professor in the NLPR Group in the Chinese Academy of Sciences, Beijing, China. Dr Böck is interested in multimodal affect processing, automatic speech processing/recognition, natural language processing, and information fusion. Current research focusses on automatic multi-party investigations,group assessments, and natural language understanding. He is organiser of several international workshops and summer schools in the fields related to human-computerinteraction and (guest) editor of multiple book and journal issues. Further, he served as Tutorial Chair (ACII 2015) and Doctorial Consortium Chair (ICHMS 2021). Dr Böck published more than 80 peer-reviewed articles and papers as well as one book.


Towards Human-Understandable XAI

Lecturer: Prof. Wojciech Samek

Date: Oct 18, 2022, 3:00 pm

Location: online


Abstract: The emerging field of Explainable AI (XAI) aims to bring transparency to today's powerful but opaque deep learning models. While local XAI methods explain individual predictions in form of attribution maps, thereby identifying “where” important features occur (but not providing information about what they represent), global explanation techniques visualize “what” concepts a model has generally learned to encode. Both types of methods thus only provide partial insights and leave the burden of interpreting the model's reasoning to the user. Building on Layer-wise Relevance Propagation (LRP), one of the most popular local XAI techniques, this talk will connect lines of local and global XAI research by introducing Concept Relevance Propagation (CRP), a next-generation XAI technique which explains individual predictions in terms of localized and human-understandable concepts. Other than the related state-of-the-art, CRP answers both the “where” and “what” question, thereby providing deep insights into the model’s reasoning process. In the talk we will demonstrate on multiple datasets, model architectures and application domains, that CRP-based analyses allow one to (1) gain insights into the representation and composition of concepts in the model as well as quantitatively investigate their role in prediction, (2) identify and counteract Clever Hans filters focusing on spurious correlations in the data, and (3) analyze whole concept subspaces and their contributions to fine-grained decision making. By lifting XAI to the concept level, CRP opens up a new way to analyze, debug and interact with ML models, which is of particular interest in safety-critical applications and the sciences.


Bio: Wojciech Samek is a professor in the Department of Electrical Engineering and Computer Science at the Technical University of Berlin and is jointly heading the Department of Artificial Intelligence at Fraunhofer Heinrich Hertz Institute (HHI), Berlin, Germany. He studied computer science at Humboldt University of Berlin, Heriot-Watt University and University of Edinburgh and received the Dr. rer. nat. degree with distinction from the Technical University of Berlin in 2014. During his studies he was awarded scholarships from the German Academic Scholarship Foundation and the DFG Research Training Group GRK 1589/1, and was a visiting researcher at NASA Ames Research Center, Mountain View, USA. Dr. Samek is associated faculty at the BIFOLD - Berlin Institute for the Foundation of Learning and Data, the ELLIS Unit Berlin and the DFG Graduate School BIOQIC, and member of the scientific advisory board of IDEAS NCBR. Furthermore, he is a senior editor of IEEE TNNLS, an editorial board member of Pattern Recognition, and an elected member of the IEEE MLSP Technical Committee. He is recipient of multiple best paper awards, including the 2020 Pattern Recognition Best Paper Award, and part of the expert group developing the ISO/IEC MPEG-17 NNR standard. He is the leading editor of the Springer book "Explainable AI: Interpreting, Explaining and Visualizing Deep Learning" (2019) and co-editor of the open access Springer book “xxAI – Beyond explainable AI” (2022). He has co-authored more than 150 peer-reviewed journal and conference papers; some of them listed as ESI Hot (top 0.1%) or Highly Cited Papers (top 1%).


Student Thesis Presentations August 2022


Date: Aug 12, 2022, 9:00 am

Location: online



  Daniel Rothenpieler


  Multimodal Video Captioning Utilising Reinforcement Learning




  Alexander Meiners


  Deep Lesion Detection Utilising Mask R-CNNs



Student Thesis Presentations March 2022


Date: Mar 30, 2022, 9:30 am

Location: online


Thomas Wagner (Bachelor): A Neural Network-Based Analysis of Facial Expressions Adaptation in Interpersonal Communication






Executive Humor: Towards a Multi-Modal Automated Measurement

Lecturer: Prof. Andreas König

Date: 11.03.2022, 15:00-16:00

Location: hybrid


Abstract: We contribute to the debate on strategic communication by leveraging machine learning for large-scale, multi-dimensional measures of executive humor. Humor, one of the most fundamental and complex phenomena in social psychology, has gathered increasing attention in management research. However, scholarly understanding of executives' humor remains substantially limited, largely because research in this domain has primarily been qualitative, survey-based, and small scale. We harness our access to a particularly suitable discursive vehicle for measuring executive humor, namely, video recordings of soccer coaches’ press conferences in the German “Bundesliga” and the English Premier League in the years 2017 to 2022. Our emerging multi-modal measure of executive humor contributes to the emerging research on humor as a key ingredient of strategic leadership as well as to computerized psychometry.



Bio: Prof. Andreas König is Chaired Professor of Strategic Management, Innovation, and Entrepreneurship at the University of Passau, Germany. He studies strategic leadership, with a specific focus on the impact of top executives' psychological characteristics on organizational outcomes and the effects of executives’ use of verbal communication on constituents’ evaluations of organizations. His research has appeared in outlets such as Administrative Science Quarterly, Academy of Management Review, and Academy of Management Journal, and he has received numerous awards, including from the Strategic Management Division of the Academy of Management, the Strategic Management Society, 2018, and the European Academy of Management. He is a member of the Editorial Board of the Academy of Management Review and Representative-at-Large at the Strategic Leadership and Governance Interest Group of the Strategic Management Society.

Interindividual differences in the relationship of emotion and performance in safety-critical systems

Lecturer: Alina Schmitz-Hübsch (University of Stuttgart, Institute of Educational Science)


Date: 15.03.2022, 14:00-15:00

Location: online


Abstract: Affect-adaptive systems detect the emotional user state, assess it against the current situation, and adjust interaction accordingly. Safety-critical systems, in which wrong decisions and behavior can have fatal consequences, may particularly benefit from affect-adaptive systems because accounting for affecting responses may help promote high performance of the human-machine-system. Effective adaptation, however, can only be accomplished when knowing which emotions benefit high performance in such systems. The results of preliminary studies indicate interindividual differences in the relationship between emotion and performance that require consideration by an affect-adaptive system. To that end, this talk introduces the concept of Affective Response Categories (ARCs) that can be used to categorize learners based on their emotion-performance relationship. In an experimental study, N = 50 subjects (33% female, 19-57 years, M = 32.75, SD = 9.8) performed a simulated airspace surveillance task. Emotional valence was detected using facial expression analysis, and pupil dilation was used to indicate emotional arousal. A cluster analysis was performed in order to group subjects into ARCs based on their individual correlations of valence and performance as well as arousal and performance. Three different clusters were identified, one of which showed no correlations between emotion and performance. The performance of subjects in all other clusters benefited from negative arousal and differed only in the valence-performance correlation, which was positive or negative. Implications for the larger context of the field of adaptive systems as well as potential benefits of the proposed concept will be discussed.


Student Thesis Presentations February 2022


Date: Feb 22, 2022, 3:00 pm

Location: online


Maximilian Spatzenegger: Acoustic Feature Analysis of Child Voice Activity: Typical Development vs. Autism Spectrum Disorder


Stefan Beil: Machine Learning-based Prediction of PDDS and PHQ-8 Scores in Persons with Multiple Sclerosis from Passive Audio Data