The Water-Futures project is an interdisciplinary project bringing together researches from machine learning, engineering and economics working on the next generation of urban drinking water systems. Among others, one objective is to develop Explainable Machine Learning models in non-stationary environments for complex structured and networked data to seamlessly support human decision making for smart water systems by data-driven technologies.

Artificial Intelligence is ubiquitous in our society and a key driver of economic growth. This comes with societal and technological challenges, as the increasing proliferation of current AI in society can reduce the autonomy of humans instead of supporting it. In order to address demands for transparency of AI, human agency, safety and resource-efficiency, the research project SAIL focuses on the full life-cycle of AI systems. This highly interdisciplinary research project (Core AI, humanities, social sciences and engineering) paves the way for systems that are sustainable with respect to short-term and long-term objectives and take care of their technical, cognitive and societal impact at reasonable operating costs.

The project ML4ProM (“Machine Learning for Process Mining”) is part of the Dataninja research training group and it aims to leverage predictive machine learning techniques to improve current efforts in process mining field. The project intends to supplement retrospective analysis with predictive Machine Learning technologies in order to comprehend the processes better. It also aims to provide predictions to enrich the information provided by the process models. Furthermore, the project would also produce alternative solutions by discovering deviations from the process model as early as possible.

The project RoSe (“Robust Individualization of Smart Sensor Technology through Transfer Learning Based Feature Selection”) is part of the Dataninja research training group and deals with the optimization of biosensor hardware and software. The hardware should be designed in such a way that the sensor is usable by many people, while it is customized for each user on the software side. Two example scenarios are considered: a novel shoe insole and a myoelectric arm orthosis.

The focus of this project is on crucial overarching properties of decisions and explanations. The aim is to investigate the important question of how a person’s critical attitude towards an AI system can be supported by fostering a healthy distrust in intelligent systems, and whether and how this attitude can be reinforced by means of explainable machine learning methods. This attitude is crucial for persons to act mindfully and be empowered to shape AI. An attitude of healthy distrust towards intelligent systems and their outputs is needed in order to prevent either disuse (e.g. a complete unwillingness to use the system) or overtrusting (e.g. blind trust) of an intelligent system. By fostering an attitude of healthy distrust a person should be able to appropriately rely on intelligent systems in potentially high-risk domains such as medicine. This project benefits from interdisciplinary expertise in experimental studies and formal modeling as well as from being able to use cases from the domain of machine learning. By combining expertise from the fields of Machine Learning and Psychology the researchers will be able to investigate the psychological underpinnings of user trust and how healthy distrust can be fostered by and in the interaction with intelligent systems, ML, and XAI.

Today, machine learning is commonly used in dynamic environments such as social networks, logistics, transportation, retail, finance, and healthcare, where new data is continuously being generated. In order to respond to possible changes in the underlying processes and to ensure that the models that have been learned continue to function reliably, they must be adapted on a continuous basis. These changes, like the model itself, should be kept transparent by providing clear explanations for users. For this, application-specific needs must be taken into account. The researchers working on Project C03 in the TRR318 are considering how and why different types of models change from a theoretical-mathematical perspective. Their goal is to develop algorithms that efficiently and reliably detect changes in models and provide intuitive explanations to users.

LeMuR is an MSCA (Marie Skłodowska-Curie Actions) Doctoral Network (DN) 2021 on Learning with Multiple Representations. The goal of LeMuR is to develop the theoretical foundations and a first set of algorithms for the new “Learning with Multiple Representations” (LMR) paradigm. Moreover, corresponding applications will be developed to demonstrate the usefulness of the new family of approaches.

Specifically, the portion of the project at Bielefeld University focuses on Learning Multiple Representations for supervised nonlinear dimensionality reduction methods by contributing to developed methods, which enable the embedding of information into low-dimensional vector spaces such that diverse and possibly changing objectives can be put into focus on-demand, which are tailored by auxiliary information such as functional properties or cognitive biases such as simplicity of the visualization and interpretability. During the course of this study, we aim to develop efficient technologies to compute such multiple embeddings efficiently and in an incremental form which is suitable for interactive exploration, couple the inference algorithm with specific domain knowledge as given in weak-supervised settings, and to evaluate efficiency and suitability in real world tasks in the medical domain, which deal with different information sources, cohorts, time scales, and attention foci.

‘Data-Driven Modelling of Metal Bending Processes’ portrays a sub-project of the priority program ‘Data-Driven Process Modelling in Metal Forming Technology’, funded by the DFG. In order to improve the quality of complex bent wire parts, it is possible to combine multi-stage mechatronic straighteners with similar bending units. However, controlling cross-stage and quantity-dependent effects of such mechatronic bending machines (MSA) represents a challenging task to experts. Machine Learning surrogate models can help to tackle this challenge by modelling defects and track data across multiple stages and piece numbers. The goal of this project is the construction of an explainable, hybrid surrogate model that allows the integration of domain knowledge such that experts have the possibility of interactively analysing accruing data of the MSA. In this project, we are collaborating with the Fraunhofer (IEM) and the University of Paderborn.

Training semantic segmentation models on multiple datasets has recently gained attention, driven by the need for robust and versatile models that can perform well across diverse visual domains. However, incompatible labelling policies between established datasets represent a major challenge which hinders principled training. In this project, we address this issue by automating the discovery of visual-semantic relations across datasets and constructing a universal taxonomy that consists of classes that describe standalone visual concepts within a dataset collection. Our approach allows us to construct models which produce predictions based on the recovered universal taxonomy and can be trained in a weakly supervised manner.

Federated learning revolutionizes machine learning by enabling collaborative training across decentralized nodes, without centralized data aggregation. Beyond privacy, it enhances system adaptability, robustness, security, and collaboration. This research leverages federated learning to optimize global systems, addressing challenges and fostering innovation in real-world distributed scenarios. With a focus on adaptability, robustness, and collaboration, the project redefines machine learning deployment in various sensitive contexts, including global energy distribution, personalized mobility, drone transportation within the Honda Research Institute EU (HRI-EU).

The Lamarr fellowship on Trustworthy AI (TrustAI) was awarded to Barbara Hammer in April 2023. The fellowship aims to support research on trustworthy artificial intelligence for spatial data to meet the challenges of climate change and drinking water supply.