SAIL

SAIL: SustAInable Life-cycle of Intelligent Socio-Technical Systems

Current systems that incorporate AI technology mainly target the introduction phase, where a core component is training and adaptation of AI models based on given example data. SAIL’s focus on the full life-cycle moves the current emphasis towards sustainable long-term development in real life. The joint project SAIL addresses both basic research in the field of AI, its implications from the perspective of the humanities and social sciences, and concrete applications in the field of Industry 4.0 and Intelligent Healthcare.
SAIL is an interdisciplinary and interinstitutional collaboration of Bielefeld University, Paderborn University, Bielefeld University of Applied Sciences, and OWL University of Applied Sciences and Arts, funded by the MKW NRW.
Below is a list of projects that are currently open and looking for applicants, please click on the link for a short description. Most projects are intended as tandems projects of PhDs, which means that two PhD candidates from different institutions and/or disciplinces can work on it. The relevant disciplinary profiles for each project are specified below; candidates are encouraged to apply to multiple projects that fit their profile. For more information, please contact the project leads mentioned.

 

 

Junior Research Groups

(R1.Ling) Longitudinal analysis of change and variety in natural language data

The PostDoc will lead a junior research group at the Faculty of Linguistics and Literature at Bielefeld University with research at the intersection of AI, linguistics, and sociology: The research focus is the development of technologies for the investigation of change of concepts in natural language, including different varieties, their representation, and their perception by humans, using social media data as data source, among others.
Disciplinary profile: You have a PhD in Linguistics, Computational Linguistics, or related areas and excellent skills in quantitative methods, statistical modeling or methods in machine learning.
Please contact Prof. Sina Zarrieß (Computational Linguistics, Bielefeld University), sina.zarriess@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

(A.6) Real-time acquisition and injection of common sense knowledge in a smart factory

According to the EU definition, AI systems in automation are high-risk AI systems and are therefore required to be bias-free. In this context, we are offering a PostDoc position who will lead a junior research group on ethical AI and cognitive assistance in automation at the Institute Industrial IT (www.init-owl.de) at OWL University of Applied Sciences and Arts conducting research at the intersection of AI, ethics and intelligent automation. The research will focus on assistance applications for multi-modal sensor systems in the context of explainable, transparent and fair AI.
Disciplinary profile: Completed PhD in computer science or a related area, strong interest in ethical aspects of AI.
Please contact Prof. Carsten Röcker (Institute Industrial IT, TH OWL), carsten.roecker@th-owl.de or Prof. Volker Lohweg (Institute Industrial IT, TH OWL), volker.lohweg@th-owl.de. For more information about the position, click here.

 

Research domain 1: Human agency to shape cooperative intelligence

(Project R1.1) Individualization of language models and language moderation

The goal is to learn how to tailor large-scale pretrained language modeling architectures to the needs of individual end users and how to automatically adapt the linguistic style of human-AI communication efficiently and meaning-preserving in continual learning. The envisioned application area is dialogue applications for healthcare.
Disciplinary profile: You have a Master’s degree in Computational Linguistics, Cognitive Science, Data Science, Computer Science, or related areas and preferably first experience in Language Models, Dialogue Models and Language Generation, Deep Learning, Expertise in Linguistics, and Expertise in Deep Learning or Reinforcement Learning.
Please contact Prof. Sina Zarrieß (Computational Linguistics, Bielefeld University), sina.zarriess@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

(Project R1.2) Eudaimonic design of work support assistance systems

As work becomes informational and demands more flexibility and judgement, the general well-being (eudaimonia) and satisfaction of basic psychological needs of the worker becomes crucial for work engagement with ITS. The goal is to understand how intelligent assistants have to be designed to balance productivity with eudaimonic objectives such as excellence, meaning and authenticity.
Disciplinary profile: You have a Master’s degree in Psychology or related areas, expertise in statistical and experimental research, as well as quantitative research, and very good knowledge in Work and Organizational Psychology.
Please contact Prof. Günter Maier (Work and Organizational Psychology, Bielefeld University), ao-psychologie@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here

 

(Project R1.3) Biases in electronic health records

Clinical notes of Electronic Health Records (EHR) are prone to biases regarding gender or ethnic stereotypes that may hinder best treatment. At the same time, it is unclear how to efficiently and reliably evaluate and mitigate bias in text, since existing measures address only specific settings. We aim to analyze effects and investigate mitigations for health records from our medical application partners on national and international scale.
Disciplinary profile: You have a Master’s degree in computer science, mathematics, cognitive science, data science, or related areas, expertise in Machine Learning, and preferably expertise in databases and statistics.
Please contact Prof. Hanna Drimalla (Multimodal signal processing, Bielefeld University), drimalla@techfak.uni-bielefeld.de. For more information about the position and application procedure for this profile, click here

 

(Project R1.5) Explaining & correcting social signals

Explainable Artificial Intelligence (XAI) currently focus mainly on explaining given categories with unique ground truth (e.g., explaining why an animal was classified as a dog/cat). In the analysis of social signals (e.g., emotional facial expression or vocal intonation), ground truth involves interpretation and uncertainty. The goal of the project is to understand individual differences in social signal expression and to develop new XAI methods for multimodal behavioral data in order to identify social signals in a transparent and correctable way. An extensive dataset of social media video data (including TikTok and YouTube) serves as the data basis for the project. The interdisciplinary project is anchored in the research group Multimodal Behavior Processing (Prof. Dr. Hanna Drimalla, Faculty of Technology) and is conducted as a cooperation with the research group Applied Social Data Science (Prof. Dr. Simon Kühne, Faculty of Sociology).
Disciplinary profile: You have a Master’s degree in Master degree in computer science, mathematics, cognitive science, data science, or related areas, expertise in Machine Learning, and preferably expertise in explainable AI and multimodal data analysis.
Please contact Prof. Hanna Drimalla (Multimodal signal processing, Bielefeld University), drimalla@techfak.uni-bielefeld.de

 

(Project R1.6) Long-term interaction memories for teachable assistance

Towards human agency in healthcare applications, it is important to develop methods that incrementally learn latent representations (encodings) allowing for efficient memory, fast recognition, and retrieval of interaction episodes with persons that have mental impairments while respecting privacy concerns. In collaboration with our application partner Bethel, this goal will be pursued using a virtual teachable personal assistant that builds up an interaction memory with a user (using natural language processing and demonstrations) and that employs it for selecting assistive behavior in related situations, respecting privacy concern.
Disciplinary profile: You have a Master’s degree in computer science, mathematics, cognitive science, data science, or related areas, expertise in Machine Learning, and preferably expertise in human-machine interaction and conversational agents.
Please contact Prof. Stefan Kopp (Cognitive Systems and Social Interaction, Bielefeld University), skopp@techfak.uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

(R1.Ling) Longitudinal analysis of change and variety in natural language data

This PhD will be carried out in a new junior research group at the Faculty of Linguistics at UNIBI with research at the intersection of AI, linguistics, and sociology: The research focus is the development of technologies for the investigation of change of concepts in natural language, including different varieties, their representation, and their perception by humans, using social media data as data source, among others.
Disciplinary profile: You have a Master’s degree in Computational Linguistics, Cognitive Science, Data Science, Computer Science, or related areas and preferably first experience in Language Models, Dialogue Models and Language Generation, Deep Learning, expertise in Computational Linguistics, quantitative methods, and preferably expertise in language models, word embeddings, language change, sociololinguistics, and experience with corpus data and corpus analysis.
Please contact Prof. Sina Zarrieß (Computational Linguistics, Bielefeld University), sina.zarriess@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

Research domain 2: Prosilience and long-term robustness through human-centered design

(Project R2.1) Self-aware AI, resilience, and preparedness

The predictions of AI systems are subject to uncertainty. This often makes their practical use difficult. There are several ways to deal with this uncertainty: These include anticipation and correction of invalid predictions, or downstream strategies for dealing with invalid results. In this project, approaches from both directions will be explored, taking into account both the algorithmic level and the user perspective. For example, AI systems interacting with the user could identify incorrect predictions and improve accordingly. The developed methods will be validated on experimental setups in the application areas “Intelligent Industrial Work Spaces” and “Adaptive Healthcare Assistance Systems”.
Disciplinary profile: You have a Master’s degree in computer science, cognitive science or related fields, basic knowledge in machine learning, very good programming skills and first experience with user surveys.
Please contact: Prof. Dr.-Ing. Wolfram Schenck (Bielefeld University of Applied Sciences), wolfram.schenck@fh-bielefeld.de

 

(Project R2.2) Data and model apoptosis

There is a right to be forgotten, but how this is conceptualized in machine learning models remains unclear so far. Conversely, how data and ML models can be archived while adhering to a set of axioms that allow reconstruction of functional and non-functional components is largely open. The goal here is to explore concepts such as the ability to reconstruct or predict events in terms of their ability to archive non-functional aspects. Central to this are questions of archiving and forgetting in the context of databases, repositories as the basis of learning processes in machine learning. The project has an interdisciplinary orientation at the interface between historical science and computer science.
Disciplinary profile: You have a Master degree in history, digital history, digital humanities, information science, data science, computer science, or related areas and you enjoy interdisciplinary work. You have good programming skills and preferably skills in statistical modeling and data science technologies.
Please contact Prof. Silke Schwandt (Digital History, Bielefeld University), silke.schwandt@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

(Project R2.3) Human-centered continuous optimization

Manual workplaces involving both humans and technical machinery are usually optimized at design time. This ignores improvements due to human learning dur- ing long-term work. Therefore, we propose to study the application of Bayesian optimization and life-long learning with a human-centered focus, e.g., a suitable weighting of past and current data or a transfer between humans.
Disciplinary profile: You have a Master’s degree in a STEM subject, and you enjoy working in interdisciplinary teams. Experience in Machine Learning and Deep Learning, excellent programming skills using Python, and experience in statistical modeling is a plus.
Please contact Prof. Barbara Hammer (Machine Learning, Bielefeld University), bhammer@techfak.uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

(Project R2.4) Robust training based on semantic adversarials

ML models are limited by design. Adversarial training aims for an increased robustness by probing models with functional errors while training. Yet, this concept is currently limited to the core technical objective of a model. We aim for its extension to semantic adversarials, which incorporate specific domain knowledge as well as run-time and hardware constraints that might be violated. Further, we elaborate on the question of how far the concept of adver sarial examples can be transferred to non-functional domains (such as biases).
Disciplinary profile 1: You have a Master’s degree in a STEM subject, and you enjoy working in interdisciplinary teams. Experience in Machine Learning and Deep Learning, excellent programming skills using Python, and experience in statistical modeling is a plus.
Please contact Prof. Barbara Hammer (Machine Learning, Bielefeld University), bhammer@techfak.uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.
Disciplinary profile 2: The person has an internationally recognized very good Master’s degree in engineering, computer science, mathematics or physics. Very good knowledge in: Artificial Intelligence, Machine Learning, Engineering Systems. The person should also have very good knowledge in Matlab and Python (Tensor Flow, PyTorch, etc.). Please contact Prof. Volker Lohweg (Information Fusion, inIT/TH OWL), volker.lohweg@th-owl.de

 

(Project R2.6) Processes of social inclusion and exclusion in hybrid teams

Hybrid teams are groups in which at least one member of the group is an autonomous technical agent. This project will investigate whether and how human members perceive the behavior of autonomous technical agents as a form of social exclusion. On this basis, the goal is to study how an enrichment of the mere functionality of intelligent technical systems by social signals and explanations of their behavior can mitigate such perception.
Disciplinary profile: You have a Master’s degree in Psychology or related areas, expertise in statistical and experimental research, as well as quantitative research, and very good knowledge in Social Psychology. Programming skills in Java are a plus.
Please contact Prof. Günter Maier (Work and Organizational Psychology, Bielefeld University), ao-psychologie@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

Research domain 3: Sustainability and efficiency in human-centered environments

(Project R3.3) Sustainable and transparent AI for small data

The aim is to develop a sustainable AI workflow that enables long-term exploitation of small data in various domains by transparent approaches. To this end, dedicated data augmentation and fitting algorithms are to be developed. The workflow is used to promote progress in many areas where only small amounts of data are available.
Disciplinary profile 1: You have a Master’s degree in a STEM subject and are highly motivated to pursue scientific questions in a collaborative environment. A background in machine learning, data analysis, Bayesian inference, optimization, stochastic processes, and the relevant implementations of software are a plus.
For questions contact Prof. Markus Lange-Hegermann (THOWL), markus.lange-hegermann@th-owl.de
Disciplinary profile 2: You have a Master’s degree in Computer Science, Engineering, Data Science or Mathematics and are motivated to work scientifically in a research network. Furthermore, if you have experience in machine learning, data science, Gaussian process models, local linear networks or XAI, and have appropriate programming skills.
Please contact Prof. Martin Kohlhase (Automation and Control, FH Bielefeld), martin.kohlhase@fh-bielefeld.de

 

(Project R3.6) Label-efficient learning from natural language supervision

The goal is to achieve robust, transparent, and – for manual annotation – efficient learning schemes by learning from approximate, real-world (natural language) task specifications rather than annotated or categorized training data.
Disciplinary profile 1: You have a Master’s degree in Computational Linguistics, Cognitive Science, Data Science, Computer Science, or related areas and preferably first experience in Language Models, Dialogue Models and Language Generation, Deep Learning, Expertise in Linguistics, and Expertise in Deep Learning or Reinforcement Learning.
Please contact Prof. Sina Zarrieß (Computational Linguistics, Bielefeld University), sina.zarriess@uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.
Disciplinary profile 2: You have a Master’s degree in a STEM subject, and you enjoy working in interdisciplinary teams. Experience in Machine Learning and Deep Learning, excellent programming skills using Python, and experience in statistical modeling is a plus.
Please contact Prof. Barbara Hammer (Machine Learning, Bielefeld University), bhammer@techfak.uni-bielefeld.de. For more information about the position and application procedure for this profile, click here.

 

Application areas: Intelligent industrial work spaces & Adaptive healthcare assistance systems

(Project A.1) Resistance to ITS in healthcare

Even though some AI-based ITS already outperform human agents in some domains of medicine, past research has identified strong resistance towards acting on information supplied by ITS in patients and healthcare providers. The project aims to answer two questions: (1) To which extent can this resistance be explained by situational factors or personality traits? (2) Which design characteristics help to overcome this resistance? We will use theories and methods from social psychology and marketing to find novel answers and overcome what might be the largest barriers to a wider implementation of AI in healthcare. The interdisciplinary project is anchored in the Psychology work group (Prof. Dr. Gerrit Hirschfeld, Faculty of Business) and will have strong connections to other SAIL-projects.

Disciplinary profile You have a Master’s degree in psychology, or related areas, expertise in empirical research methods, and preferably expertise in designing, conducting and analyzing experiments in social psychology.
Please contact: Prof. Dr. Gerrit Hirschfeld (Bielefeld University of Applied Sciences), gerrit.hirschfeld@fh-bielefeld.de

 

(Project A.4) Care bed robotics

There are various life situations in which people depend on care in a nursing bed in the course of or as a consequence of an illness. Nursing beds, especially those for intensive care with a large number of actuators, can themselves be seen as robots that take care of the users in them, e.g. by performing positioning movements to prevent bedsores. In the present project, a nursing bed is available that has RGB, depth imaging and infrared cameras as well as sensors for pressure distribution measurements in the lying surface and whose actuators can be controlled by computer. The project starts with a fusion of the sensor data and uses ML-based modelling approaches to determine/estimate process variables and patient characteristics that cannot be measured directly from a subset of the sensors (e.g. estimation of the lying pressure distribution without pressure distribution measurement - as a basis for bed positioning movements). Subsequent AI questions deal, for example, with model individualisation, natural interaction modes and adaptivity.

Disciplinary profile: You have a Master’s degree in (Technical) Computer Science, Electrical Engineering, (Bio-) Mechatronics or related fields, expertise in acquisition, processing and algorithmic evaluation of (image) sensor data, basic expertise in the field of machine learning and in dealing with embedded (robotic) systems.
Please contact: Prof. Dr. Axel Schneider (Bielefeld University of Applied Sciences), axel.schneider@fh-bielefeld.de