Publications

LOLA – An Open-Source Massively Multilingual Large Language Model

Authors: Nikit Srivastava, Denis Kuchelev, Tatiana Moteu Ngoli, Kshitij Shetty, Michael Röder, Hamada Zahera, Diego Moussallem, Axel-Cyrille Ngonga Ngomo

This paper presents LOLA, a massively multilingual large language model trained on more than 160 languages using a sparse Mixture-of-Experts Transformer architecture. Our architectural and implementation choices address the challenge of harnessing linguistic diversity while maintaining efficiency and avoiding the common pitfalls of multilinguality. Our analysis of the evaluation results shows competitive performance in natural language generation and understanding tasks. (...)

Contextual Augmentation for Entity Linking using Large Language Models

Authors: Daniel Vollmers, Hamada M. Zahera, Diego Moussallem, Axel-Cyrille Ngonga Ngomo

Entity Linking involves detecting and linking entity mentions in natural language texts to a knowledge graph. Traditional methods use a two-step process with separate models for entity recognition and disambiguation, which can be computationally intensive and less effective. We propose a fine-tuned model that jointly integrates entity recognition and disambiguation in a unified framework. (...)

Open challenges for the automatic synthesis of clinical trials

Authors: Olivia Sanchez-Graillet, David M. Schmidt, Christian Kullik & Philipp Cimiano

An important criterion for selecting clinical trials to be compared in systematic reviews and meta-analyses is that they measure the same outcomes. However, this represents a challenge as there is a wide variety of outcomes, and it is difficult to standardize them for comparing clinical trials containing them. To address this challenge, we utilized our annotated dataset, which includes 211 abstracts of clinical trials related to glaucoma and type 2 diabetes mellitus. We then developed a tool that provides an overview of the annotated clinical trial information and enables users to group them by outcomes. (...)

Grid-Oriented Control of Vehicle Batteries in a Cellular Grid Setup Based on Fuzzy Logic

Authors: Lars Quakernack, Melina Gurcke, Katrin Schulte, Jens Haubrock

The electrification of various sectors and the expansion of renewable energy resources (RES) leads to a change from the historically established and planned vertical load flow in the electrical power system to a horizontal one. This is placing a particular strain on the distribution grids to which the new loads and decentralized generators are connected. The cellular energy system approach is expected to work with a high proportion of RES and ensure a high level of supply security. This paper investigates the autonomous control of vehicle batteries for a cellular grid approach. (...)

LSTM Autoencoder Model to Recognize Electric Vehicles in Grouped Smart Meter Data

Authors: Lars Quakernack, Thomas Engelmann, Jens Haubrock, Valerie Vaquet

Uncertainty in controllable devices and their power in distribution grids is a considerable problem for grid operators. The corresponding "blind" control of electric vehicles (EV), heat pumps, heating, ventilation, and air conditioning systems can harm the grid. On the one hand, if not enough controllable devices are available to balance the load, congestion, potentially damaging the operating equipment, can occur. On the other hand, the incentive of prosumer involvement to provide flexibility can decrease due to overcontrol of their devices. (...)

Substituting confidence for competence in health literacy: a review of studies, citations, and trial registrations

Authors: Inga Jagemann, Christian Thiele, Ruth von Brachel, Gerrit Hirschfeld

Patient health literacy is crucial for effective patient–physician communication, and interventions targeting health literacy can use measures based on either actual performance (competence) or self-ratings (confidence). This paper analyzed the development of these measures through three studies. Study 1 reviewed articles describing the development of novel measures; Study 2 examined the citations of these studies, and Study 3 evaluated data from clinical trials registries. (...)

Equivariance and partial observations in Koopman operator theory for partial differential equations

Authors: Sebastian Peitz, Hans Harder, Feliks Nüske, Friedrich M Philipp, Manuel Schaller, Karl Worthmann

The Koopman operator has become an essential tool for data-driven analysis, prediction, and control of complex systems. The main reason is the enormous potential of identifying linear function space representations of nonlinear dynamics from measurements. This equally applies to ordinary, stochastic, and partial differential equations (PDEs). (...)

Acceptance of Medical Artificial Intelligence in Skin Cancer Screening: Choice-Based Conjoint Survey

Authors: Inga Jagemann, Ole Wensing, Manuel Stegemann, Gerrit Hirschfeld

There is great interest in using artificial intelligence (AI) to screen for skin cancer. This is fueled by a rising incidence of skin cancer and an increasing scarcity of trained dermatologists. AI systems capable of identifying melanoma could save lives, enable immediate access to screenings, and reduce unnecessary care and health care costs. While such AI-based systems are useful from a public health perspective, past research has shown that individual patients are very hesitant about being examined by an AI system.

Evaluation of Swin Transformer and knowledge transfer for denoising of super-resolution structured illumination microscopy data

Authors: Zafran Hussain Shah, Marcel Müller, Wolfgang Hübner, Tung-Cheng Wang, Daniel Telman, Thomas Huser, Wolfram Schenck

Convolutional neural network (CNN)–based methods have shown excellent performance in denoising and reconstruction of super-resolved structured illumination microscopy (SR-SIM) data. Therefore, CNN-based architectures have been the focus of existing studies. However, Swin Transformer, an alternative and recently proposed deep learning–based image restoration architecture, has not been fully investigated for denoising SR-SIM images. (...)

A Sensor Fault Detection and Imputation Framework for Electrical Distribution Grids

Authors: Lars Quakernack, Valerie Vaquet, Barbara Hammer, Jens Haubrock

Automated and smart methods for monitoring and controlling the low voltage grid are required in the future to ensure safe operations in the presence of increasingly fluctuating power generation caused by distributed energy resources and power peaks caused by a rising number of electrical vehicles. These algorithmic methods rely on accurate (real-time) data (...)

Active Learning for Handling Missing Data

Authors: Alaa Tharwat, Wolfram Schenck

Recently, the massive growth of IoT devices and Internet data, which are widely used in many applications, including industry and healthcare, has dramatically increased the amount of free unlabeled data collected. However, this unlabeled data is useless if we want to learn supervised machine learning models. The expensive and time-consuming cost of labeling makes the problem even more challenging. Here, the active learning (AL) technique provides a solution (...)

Distributed control of partial differential equations using convolutional reinforcement learning

Authors: Sebastian Peitz, Jan Stenner , Vikas Chidananda, Oliver Wallscheid, Steven L. Brunton, Kunihiko Taira

We present a convolutional framework which significantly reduces the complexity and thus, the computational effort for distributed reinforcement learning control of dynamical systems governed by partial differential equations (PDEs). Exploiting translational equivariances, the high-dimensional distributed control problemcan be transformed into a multi-agent control problem with many identical, uncoupled agents. (...)

Using methods from dimensionality reduction for active learning with low query budget

Authors: Alaa Tharwat, Wolfram Schenck

Recently, it has been challenging to generate enough labeled data for supervised learning models from a large amount of free unlabeled data due to the high cost of the labeling process. Here, the active learning technique provides a solution by annotating a small but highly informative set of unlabeled data. This ensures high generalizability in space and improves classification performance with test data. The task is more challenging when (...)

EGNN-C+: Interpretable Evolving Granular Neural Network and Application in Classification of Weakly-Supervised EEG Data Streams

Authors: Daniel Leite, Alisson SIlva, Gabriella Casalino, Arnab Sharma, Danielle Fortunato, Axel-Cyrille Ngonga Ngomo

We introduce a modified incremental learning algorithm for evolving Granular Neural Network Classifiers (eGNNC+). We use double-boundary hyper-boxes to represent granules, and customize the adaptation procedures to enhance the robustness of outer boxes for data coverage and noise suppression, while ensuring that inner boxes remain flexible to capture drifts. The classifier evolves from scratch, incorporates new classes on the fly, and performs local incremental feature weighting. As an application, we focus on the classification of emotion-related patterns within electroencephalogram (EEG) signals. Emotion recognition is crucial (...)

Efficiently Computable Safety Bounds for Gaussian Processes in Active Learning

Authors: Jörn Tebbe, Christoph Zimmer, Ansgar Steland, Markus Lange-Hegermann, Fabian Mies

Active learning of physical systems must commonly respect practical safety constraints, which restricts the exploration of the design space. Gaussian Processes (GPs) and their calibrated uncertainty estimations are widely used for this purpose. In many technical applications the design space is explored via continuous trajectories, along which the safety needs to be assessed. This is particularly challenging for strict safety requirements in GP methods, as it employs computationally expensive Monte-Carlo sampling of high quantiles. We address these challenges by providing (...)

On the continuity and smoothness of the value function in reinforcement learning and optimal control

Authors: Hans Harder, Sebastian Peitz

he value function plays a crucial role as a measure for the cumulative future reward an agent receives in both reinforcement learning and optimal control. It is therefore of interest to study how similar the values of neighboring states are, i.e., to investigate the continuity of the value function. We do so by providing and verifying upper bounds on the value function's modulus of continuity. Additionally, we show that the value function is always (...)

Comparing generative and extractive approaches to information extraction from abstracts describing randomized clinical trials

Authors: Christian Witte, David M. Schmidt, Philipp Cimiano

Systematic reviews of Randomized Controlled Trials (RCTs) are an important part of the evidence-based medicine paradigm. However, the creation of such systematic reviews by clinical experts is costly as well as time-consuming, and results can get quickly outdated after publication. Most RCTs are structured based on the Patient, Intervention, Comparison, Outcomes (PICO) framework and there exist many approaches which aim to extract PICO elements automatically. (...)

Predicting PDEs Fast and Efficiently with Equivariant Extreme Learning Machines

Authors: Hans Harder, Sebastian Peitz

We utilize extreme learning machines for the prediction of partial differential equations (PDEs). Our method splits the state space into multiple windows that are predicted individually using a single model. Despite requiring only few data points (in some cases, our method can learn from a single full-state snapshot), it still achieves high accuracy and can predict the flow of PDEs over long time horizons. Moreover, we show how additional symmetries can be exploited to increase sample efficiency and to enforce equivariance. (...)

Foundation Model Vision Transformers are Great Tracking Backbones

Authors: Tristan Kenneweg, Philp Kenneweg, Barbara Hammer

The recent breakthroughs in foundation models for image processing [1], [2] have made using Vision Transformer embeddings for downstream image tasks a great option in many applications. However, best practices on how to use these embeddings for a given task have not been established yet.In this paper we investigate the suitability of foundation models for the single object tracking task. We do this by developing and implementing a zero-shot patch tracking method and a deep learning system which builds upon foundation model Vision Transformer embeddings. We evaluate these methods (...)

Methods for Estimating the Detection and Quantification Limits of Key Substances in Beer Maturation with Electronic Noses

Authors: Julia Kruse, Julius Wörner, Jan Schneider, Helene Dörksen, Miriam Pein-Hackelbusch

o evaluate the suitability of an analytical instrument, essential figures of merit such as the limit of detection (LOD) and the limit of quantification (LOQ) can be employed. However, as the definitions k nown in the literature are mostly applicable to one signal per sample, estimating the LOD for substances with instruments yielding multidimensional results like electronic noses (eNoses) is still challenging. In this paper, we will compare and present different approaches to estimate the LOD for eNoses by employing (...)

Efficient Evaluation of Conjunctive Regular Path Queries Using Multi-way Joins

Authors: Nikolaos Karalis, Alexander Bigerl, Liss Heidrich, Mohamed Ahmed Sherif, Axel-Cyrille Ngonga Ngomo

Recent analyses of real-world queries show that a prominent type of queries is that of conjunctive regular path queries. Despite the increasing popularity of this type of queries, only limited efforts have been invested in their efficient evaluation. Motivated by recent results on the efficiency of worst-case optimal multi-way join algorithms for the evaluation of conjunctive queries, we present a novel multi-way join algorithm for the efficient evaluation of conjunctive regular path queries. (...)

Universal Knowledge Graph Embeddings

Authors: N’Dah Jean Kouago, Caglar Demir, Hamada M. Zahera, Adrian Wilke, Stefan Heindorf, Jiayi Li, Axel-Cyrille Ngonga Ngomo

A variety of knowledge graph embedding approaches have been developed. Most of them obtain embeddings by learning the structure of the knowledge graph within a link prediction setting. As a result, the embeddings reflect only the structure of a single knowledge graph, and embeddings for different knowledge graphs are not aligned, e.g., they cannot be used to find similar entities across knowledge graphs via nearest neighbor search. However, knowledge graph embedding applications such as entity disambiguation require a more global representation, i.e., a representation that is valid across multiple sources. We propose to learn universal knowledge graph embeddings (...)

Active Learning für Regressionsprobleme mit Ensemble-Methoden – Untersuchung des Trade-Offs zwischen Qualität und Rechenaufwand

Authors: Bjarne Jaster, Martin Kohlhase

Uncertainty estimators are often used for Active Learning (Uncertainty sampling). Depending on the base-model the quality and the computational effort of these estimators can vary. This work explores this trade-off for three different base-models, that are used to build ensembles to get the uncertainty estimates.

A Hybrid Spiking-Convolutional Neural Network Approach for Advancing Machine Learning Models

Authors: Sanaullah Sanaullah, Kaushik Roy, Ulrich Rückert, Thorsten Jungeblut

In this article, we propose a novel standalone hybrid Spiking-Convolutional Neural Network (SC-NN) model and test on using image inpainting tasks. Our approach uses the unique capabilities of SNNs, such as event-based computation and temporal processing, along with the strong representation learning abilities of CNNs, to generate high-quality inpainted images. The model is trained (...)

WikiScenes with Descriptions: Aligning Paragraphs and Sentences with Images in Wikipedia Articles

Authors: Özge Alaçam, Ronja Utescher, Hannes Grönner, Judith Sieker, Sina Zarrieß

Research in Language & Vision rarely uses naturally occurring multimodal documents as Wikipedia articles, since they feature complex image-text relations and implicit image-text alignments. In this paper, we provide one of the first datasets that provides ground-truth annotations of image-text alignments in multi-paragraph multi-image articles. The dataset can be used to study phenomena of visual languag (...)

How turn-timing can inform about becoming familiar with a task and its changes: a study of shy and less shy four-year-old children

Authors: Valeriia Tykhonenko, Nils F. Tolksdorf, Katharina Rohlfing

In novel situations, the productive communicative behavior of shy children can require more time than that of their less shy peers. Investigating 14 preschoolers, we asked which situational demands and changes contribute to the individual processing. Whereas children’s shyness was measured by a standardized questionnaire given to caregivers, their processing of situational demands was measured by their nonverbal turn-timing over two sessions with a social robot. We focused on (...)

Performance without understanding: How ChatGPT relies on humans to repair conversational trouble

Authors: Ole Pütz, Elena Esposito

LLM-based chatbots’ ability to generate contextually appropriate and informative texts can be taken as an indication that they are also able to understand text. We argue instead that the separation of the two competences to generate and to understand text is the key to their performance in dialog with human users. This argument requires a shift in perspective from a concern with machine intelligence to a concern with communicative competence. We illustrate our argument with empirical examples (...)

Benchmarking Low-Resource Machine Translation Systems

Authors: Ana Silva, Nikit Srivastava, Tatiana Moteu Ngoli, Michael Röder, Diego Moussallem, Axel-Cyrille Ngonga Ngomo

ssessing the performance of machine translation systems is of critical value, especially to languages with lower resource availability.Due to the large evaluation effort required by the translation task, studies often compare new systems against single systems or commercial solutions. Consequently, determining the best-performing system for specific languages is often unclear. This work benchmarks publicly available translation systems across 4 datasets and 26 languages, including low-resource languages. We consider (...)

Evaluating Task-Level Struggle Detection Methods in Intelligent Tutoring Systems for Programming

Authors: Jesper Dannath, Alina Deriyeva, Benjamin Paaßen

Intelligent Tutoring Systems require student modeling in order to make pedagogical decisions, such as individualized feedback or task selection. Typically, student modeling is based on the eventual correctness of tasks. However, for multi-step or iterative learning tasks, like in programming, the intermediate states towards a correct solution also carry crucial information about learner skill. We investigate how to detect learners who struggle on their path towards a correct solution of a task. (...)

Relation between struggle and learning personality in programming exercises

Authors: Alina Deriyeva, Jesper Dannath, Benjamin Paaßen

Personality-related characteristics can have an impact on learning experiences and learning outcomes. Moreover, understanding learning approaches of students can help to make personalized pedagogical decisions. Particularly, this has a potential to improve learning outcomes and mitigate user attrition in digital learning environments (DLEs). We hypothesize that persistent individual characteristics may influence a learners’ tendency to struggle during programming exercises. In a study (...)

Revolutionizing Qualitative Human-Robot Interaction Research by Using GPT Models for Inductive Category Development

Authors: Clarissa Sabrina Arlinghaus, Charlotte Wulff, Günter W. Maier

Coding qualitative data is essential but time-consuming. This late-breaking report presents a new method for developing inductive categories utilizing GPT models. We examined two different GPT models (gpt-3.5-turbo-0125 and gpt-4o-2024-05-03) and three temperature settings (0, 0.5, 1), each with ten repetitions. The generated categories were fairly consistent across settings, although higher temperatures included less relevant aspects. (...)

A Robot’s Moral Advice Is Not Appreciated Neither in Functional nor in Social Communication

Authors: Clarissa Sabrina Arlinghaus, Carolin Straßmann, Annika Dix

This study (N = 317) investigated the influence of verbal communication (social vs. functional) on the acceptance of robot recommendations in non-moral, somewhat moral or very moral decision-making situations. The robot’s communication style had no impact on the participants (1) being confident in their decision, (2) perceiving the robot’s recommendation as helpful, and (3) making a decision dependent on the robot’s recommendation. (...)

EDGE: Evaluation Framework for Logical vs. Subgraph Explanations for Node Classifiers on Knowledge Graphs

Authors: Rupesh Sapkota, Dominik Köhler, Stefan Heindorf

As machine learning and deep learning become increasingly integrated into our daily lives, understanding how these technologies make decisions is crucial. To ensure transparency, accountability, and ethical adherence, these so-called “black-box” models should be accompanied by human-comprehensible explanations of their predictions. This clarity is essential for establishing trust in their real-world applications. (...)

The SAME score: Improved cosine based bias score for word embeddings

Authors: Sarah Schröder, Alexander Schulz, Barbara Hammer

With the enourmous popularity of large language models, many researchers have raised ethical concerns regarding social biases incorporated in such models. Several methods to measure social bias have been introduced, but apparently these methods do not necessarily agree regarding the presence or severity of bias. Furthermore, some works have shown theoretical issues or severe limitations with certain bias measures. (...)

Semantic Properties of cosine based bias scores for word embeddings

Authors: Sarah Schröder, Alexander Schulz, Fabian Hinder, Barbara Hammer

Plenty of works have brought social biases in language models to attention and proposed methods to detect such biases. As a result, the literature contains a great deal of different bias tests and scores, each introduced with the premise to uncover yet more biases that other scores fail to detect. What severely lacks in the literature, however, are comparative studies that analyse such bias scores and help researchers to understand the benefits or limitations of the existing methods. (...)

Koopman-Based Surrogate Modelling of Turbulent Rayleigh-Bénard Convection

Authors: Thorben Markmann, Michiel Straat, Barbara Hammer

Several related works have introduced Koopman-based Machine Learning architectures as a surrogate model for dynamical systems. These architectures aim to learn non-linear measurements (also known as observables) of the system’s state that evolve by a linear operator and are, therefore, amenable to model-based linear control techniques. So far, mainly simple systems have been targeted, and Koopman architectures as reduced-order models for more complex dynamics have not been fully explored. (...)

ROCES: Robust Class Expression Synthesis in Description Logics via Iterative Sampling

Authors: N’Dah Jean Kouagou , Stefan Heindorf, Caglar Demir, Axel-Cyrille Ngonga Ngomo

We consider the problem of class expression learning using cardinality-minimal sets of examples. Recent class expression learning approaches employ deep neural networks and have demonstrated tremendous performance improvements in execution time and quality of the computed solutions. However, they lack generalization capabilities when it comes to the number of examples used in a learning problem, i.e., they often perform poorly on unseen learning problems where only a few examples are given. In this work, we propose a generalization of the classical class expression learning problem to address the limitations above. (...)

Image restoration in frequency space using complex-valued CNNs

Authors: Zafran Hussain Shah, Marcel Müller, Wolfgang Hübner, Henning Ortkrass, Barbara Hammer, Thomas Huser, Wolfram Schenck

Real-valued convolutional neural networks (RV-CNNs) in the spatial domain have outperformed classical approaches in many image restoration tasks such as image denoising and super-resolution. Fourier analysis of the results produced by these spatial domain models reveals the limitations of these models in properly processing the full frequency spectrum. This lack of complete spectral information can result in missing textural and structural elements. To address this limitation, we explore the potential of complex-valued convolutional neural networks (CV-CNNs) for image restoration tasks. (...)

Exploration Techniques in Active Learning in Classification

Authors: Peter Kuchling

Active Learning is the process of selectively querying unlabelled data to be classified by an expert for efficient supervised learning on small data to, among other things, reduce labelling costs. While many works are concerned with the notion of uncertainty quantification as a measure to query informative data points, less work has been dedicated to the preceding exploration phase. (...)

Evaluating Negation with Multi-way Joins Accelerates Class Expression Learning

Authors: Nikolaos Karalis, Alexander Bigerl, Caglar Demir, Liss Heidrich, Axel-Cyrille Ngonga Ngomo

Class expression learning based on refinement operators is a popular family of explainable machine learning approaches for RDF knowledge graphs with ontologies in description logics. However, most implementations of this paradigm fail to scale to the large knowledge graphs found on the Web. One common bottleneck of these implementations is the instance retrieval function. We address this drawback by introducing an algorithm inspired by worst-case optimal multi-way joins for the evaluation of SPARQL queries (...)

Generating SPARQL from Natural Language Using Chain-of-Thoughts Prompting

Authors: Hamada M. Zahera, Manzoor Ali, Mohamed Ahmed Sherif, Diego Mousallem, Axel-Cyrille Ngonga Ngomo

Purpose: SPARQL is a highly expressive query language for knowledge graphs; yet, formulating precise SPARQL queries can be challenging for non-expert users. A potential solution is translating natural questions into SPARQL queries, known as SPARQL generation. This paper addresses the challenges of translating natural language questions into SPARQL queries for different knowledge graphs. (...)

BLINK: Blank Node Matching Using Embeddings

Authors: Alexander Becker, Mohamed Ahmed Sherif, Axel-Cyrille Ngonga Ngomo

Knowledge graphs (KGs) differ significantly over multiple different versions of the same data source. They also often contain blank nodes that do not have a constant identifier over all versions. Linking such blank nodes from different versions is a challenging task. Previous works propose different approaches to create signatures for all blank nodes based on named nodes in their neighborhood to match blank nodes with similar signatures (...)

What you need to know about a learning robot: Identifying the enabling architecture of complex systems

Authors: Helen Beierling, Phillip Richter, Mara Brandt, Lutz Terfloth, Carsten Schulte, Heiko Wersing, Anna-Lisa Vollmer

Nowadays we deal with robots and AI more and more in our everyday life. However, their behavior is not always apparent to most lay users, especially in error situations. This can lead to misconceptions about the behavior of the technologies being used. This in turn can lead to misuse and rejection by users. Explanation, for example through transparency, can address these misconceptions. (...)

Gender differences in preferences for mental health apps in the general population – a choice-based conjoint analysis from Germany

Authors: Inga Jagemann, Manuel Stegemann, Ruth von Brachel, Gerrit Hirschfeld

Background: Men and women differ in the mental health issues they typically face. This study aims to describe gender differences in preferences for mental health treatment options and specifically tries to identify participants who prefer AI-based therapy over traditional face-to-face therapy. Method: A nationally representative sample of 2,108 participants (53% female) aged 18 to 74 years completed a choice-based conjoint analysis (CBCA). Within the CBCA, participants evaluated twenty choice sets, each describing three treatment variants in terms of provider, content, costs, and waiting time. (...)

Group-Convolutional Extended Dynamic Mode Decomposition

Authors: Hans Harder, Feliks Nüske, Friedrich M. Philipp, Manuel Schaller, Karl Worthmann, Sebastian Peitz

This paper explores the integration of symmetries into the Koopman-operator framework for the analysis and efficient learning of equivariant dynamical systems using a group-convolutional approach. Approximating the Koopman operator by finite-dimensional surrogates, e.g., via extended dynamic mode decomposition (EDMD), is challenging for high-dimensional systems due to computational constraints. To tackle this problem with a particular focus on EDMD, we demonstrate -- under suitable equivarance assumptions on the system and the observables -- that the optimal EDMD matrix is equivariant. (...)

ExPrompt: Augmenting Prompts Using Examples as Modern Baseline for Stance Classification

Authors: Umair Qudus, Michael Röder, Daniel Vollmers, Axel-Cyrille Ngonga Ngomo

Detecting the veracity of a statement automatically is a challenge the world is grappling with due to the vast amount of data spread across the web. Verifying a given claim typically entails validating it within the framework of supporting evidence like a retrieved piece of text. Classifying the stance of the text with respect to the claim is called stance classification. (...)

BLINK: Blank Node Matching Using Embeddings

Authors: Alexander Becker, Mohamed Ahmed Sherif, Axel-Cyrille Ngonga Ngomo

Knowledge graphs (KGs) differ significantly over multiple different versions of the same data source. They also often contain blank nodes that do not have a constant identifier over all versions. Linking such blank nodes from different versions is a challenging task. Previous works propose different approaches to create signatures for all blank nodes based on named nodes in their neighborhood to match blank nodes with similar signatures. (...)

Hateful Word in Context Classification

Authors: Sanne Hoeken, Sina Zarrieß, Özge Alacam

Hate speech detection is a prevalent research field, yet it remains underexplored at the level of word meaning. This is significant, as terms used to convey hate often involve non-standard or novel usages which might be overlooked by commonly leveraged LMs trained on general language use. In this paper, we introduce the Hateful Word in Context Classification (HateWiC) task and present a dataset of ~4000 WiC-instances, each labeled by three annotators. (...)

Eyes Don’t Lie: Subjective Hate Annotation and Detection with Gaze

Authors: Özge Alacam, Sanne Hoeken, Sina Zarrieß

Hate speech is a complex and subjective phenomenon. In this paper, we present a dataset (GAZE4HATE) that provides gaze data collected in a hate speech annotation experiment. We study whether the gaze of an annotator provides predictors of their subjective hatefulness rating, and how gaze features can improve Hate Speech Detection (HSD). (...)

The Illusion of Competence: Evaluating the Effect of Explanations on Users’ Mental Models of Visual Question Answering Systems

Authors: Judith Sieker, Simeon Junker, Ronja Utescher, Nazia Attari, Heiko Wersing, Hendrik Buschmeier, Sina Zarrieß

We examine how users perceive the limitations of an AI system when it encounters a task that it cannot perform perfectly and whether providing explanations alongside its answers aids users in constructing an appropriate mental model of the system’s capabilities and limitations. (...)

Hidden in Plain Sight: Adversarial Attack on Wavelet-Based Banknote Authentication

Authors: Julian Knaup, Christoph-Alexander Holst, Volker Lohweg

Machine learning systems are increasingly integrated into security-relevant applications, making their vul-nerability to adversarial examples a potential risk. Banknote authentication is one such use case that ensures trustworthiness in financial transactions. In addition to traditional security features, the printing technique of banknotes itself is leveraged for authentication. The Intaglio printing produces particularly fine print structures that can be analyzed and differentiated using spatial frequency analysis, e.g. the wavelet packet transform. (...)

Fine-Grained Detection of Solidarity for Women and Migrants in 155 Years of German Parliamentary Debates

Authors: Aida Kostikova, Benjamin Paassen, Dominik Beese, Ole Pütz, Gregor Wiedemann, Steffen Eger

Solidarity is a crucial concept to understand social relations in societies. In this study, we investigate the frequency of (anti-)solidarity towards women and migrants in German parliamentary debates between 1867 and 2022. Using 2,864 manually annotated text snippets, we evaluate large language models (LLMs) like Llama 3, GPT-3.5, and GPT-4. We find that GPT-4 outperforms other models, approaching human annotation accuracy. Using GPT-4, we automatically annotate 18,300 further instances and find that solidarity with migrants outweighs anti-solidarity but that frequencies and solidarity types shift over time. (...)

FaVEL: Fact Validation Ensemble Learning

Authors: Umair Qudus, Franck Lionel Tatkeu Pekarou, Ana Alexandra Morim da Silva, Michael Röder, Axel-Cyrille Ngonga Ngomo

Validating assertions before adding them to a knowledge graph is an essential part of its creation and maintenance. Due to the sheer size of knowledge graphs, automatic fact-checking approaches have been developed. These approaches rely on reference knowledge to decide whether a given assertion is correct. Recent hybrid approaches achieve good results by including several knowledge sources. However, it is often impractical to provide a sheer quantity of textual knowledge or generate embedding models to leverage these hybrid approaches. We present FaVEL, an approach that uses algorithm selection (...)

Case study: Using LLMs to assist with solving programming homework assignments

Authors: Alina Deriyeva, Jesper Dannath, Benjamin Paaßen

Nowadays, students have the option of using LLMs for assistance in solving homework assignments. Moreover, most LLMs, like ChatGPT, are also trained on large sets of source code and thus can be used to assist in programming exercises. In this paper, we present a case study based on data collected over the course of 1.5 semesters, where students of three programming-related courses were explicitly permitted to use such models while solving homework assignments. (...)

Interpretability Index Based on Balanced Volumes for Transparent Models and Agnostic Explainers

Authors: Daniel Leite , Arnab Sharma, Caglar Demir, Axel-Cyrille Ngonga Ngomo

We discuss interpretability and explainability of machine learning models. We introduce a universal interpretability index, JJ, to quantify and monitor the interpretability of a general-purpose model, which can be static or evolve incre-mentally from a data stream. The models can be transparent classifiers, predictors or controllers operating on partitions or granules of the data space, e.g., rule-based models, trees, proba-bilistic clustering models, modular or granular neural networks. (...)

Trading-Off Interpretability and Accuracy in Medical Applications: A Study Toward Optimal Explainability of Hoeffding Trees

Authors: Arnab Sharma, Daniel Leite, Caglar Demir, Axel-Cyrille Ngonga Ngomo

With an increased number of applications of machine learning models to support decision-making in critical domains, there is a pressing need to understand the internal behavior of these models. Essentially, explaining learning models to humans has expedited the development of methods to extract information and access models' inner components. Researchers have proposed approaches to compute explanations for different types of models.(...)

Learning a model is paramount for sample efficiency in reinforcement learning control of PDEs

Authors: Stefan Werner, Sebastian Peitz

The goal of this paper is to make a strong point for the usage of dynamical models when using reinforcement learning (RL) for feedback control of dynamical systems governed by partial differential equations (PDEs). To breach the gap between the immense promises we see in RL and the applicability in complex engineering systems, the main challenges are the massive requirements in terms of the training data, as well as the lack of performance guarantees. We present a solution(...)

Learning Permutation-Invariant Embeddings for Description Logic Concepts

Authors: Caglar Demir, Axel-Cyrille Ngonga Ngomo

Concept learning deals with learning description logic concepts from a background knowledge and input examples. The goal is to learn a concept that covers all positive examples, while not covering any negative examples. This non-trivial task is often formulated as a search problem within an infinite quasi-ordered concept space. Although state-of-the-art models have been successfully applied to tackle this problem, their large-scale applications have been severely hindered due to their excessive exploration incurring impractical runtimes. Here, we propose a remedy for this limitation. (...)

Enhancing Comprehension and Navigation in Jupyter Notebooks with Static Analysis

Authors: Ashwin Prasad Shivarpatna Venkatesh, Jiawei Wang, Li Li, Eric Bodden

Jupyter notebooks enable developers to interleave code snippets with rich-text and in-line visualizations. Data scientists use Jupyter notebook as the de-facto standard for creating and sharing machine-learning based solutions, primarily written in Python. Recent studies have demonstrated, however, that a large portion of Jupyter notebooks available on public platforms are undocumented and lacks a narrative structure. This reduces the readability of these notebooks. To address this shortcoming, this paper presents HeaderGen (...)

A Survey on Active Learning: State-of-the-Art, Practical Challenges and Research Directions

Authors: Alaa Tharwat, Wolfram Schenck

Despite the availability and ease of collecting a large amount of free, unlabeled data, the expensive and time-consuming labeling process is still an obstacle to labeling a sufficient amount of training data, which is essential for building supervised learning models. Here, with low labeling cost, the active learning (AL) technique could be a solution (...)

Unsupervised Cyclic Siamese Networks Automating Cell Imagery Analysis

Authors: Dominik Stallmann, Barbara Hammer

Novel neural network models that can handle complex tasks with fewer examples than before are being developed for a wide range of applications. In some fields, even the creation of a few labels is a laborious task and impractical, especially for data that require more than a few seconds to generate each label. In the biotechnological domain, cell cultivation experiments are usually done by varying the circumstances of the experiments (...)

Identifying Slurs and Lexical Hate Speech via Light-Weight Dimension Projection in Embedding Space

Authors: Sanne Hoeken, Sina Zarrieß, Özge Alacam

The prevalence of hate speech on online platforms has become a pressing concern for society, leading to increased attention towards detecting hate speech. Prior work in this area has primarily focused on identifying hate speech at the utterance level that reflects the complex nature of hate speech. In this paper, we propose a targeted and efficient approach to identifying hate speech by detecting slurs at the lexical level using contextualized word embeddings. We hypothesize that slurs (...)

A Unifying Formal Approach to Importance Values in Boolean Functions

Authors: Hans Harder, Simon Jantsch, Christel Baier, Clemens Dubslaff

Boolean functions and their representation through logics, circuits, machine learning classifers, or binary decision diagrams (BDDs) play a central role in the design and analysis of computing systems. Quantifying the relative impact of variables on the truth value by means of importance values can provide useful insights to steer system design and debugging. In this paper, we introduce a uniform framework for reasoning about such value (...)

Neural Class Expression Synthesis

Authors: N’Dah Jean Kouagou, Stefan Heindorf, Caglar Demir, Axel-Cyrille Ngonga Ngomo

Many applications require explainable node classification in knowledge graphs. Towards this end, a popular “white-box” approach is class expression learning: Given sets of positive and negative nodes, class expressions in description logics are learned that separate positive from negative nodes. Most existing approaches are search-based approaches generating many candidate class expressions and selecting the best one. However, they often take a long time to find suitable class expressions. In this paper, we cast class expression learning (...)

Explainable Integration of Knowledge Graphs using Large Language Models

Authors: Abdullah Fathi Ahmed, Asep Fajar Firmansyah, Mohamed Ahmed Sherif, Diego Moussallem, Axel-Cyrille Ngonga Ngomo

Linked knowledge graphs build the backbone of many data-driven applications such as search engines, conversational agents and e-commerce solutions. Declarative link discovery frameworks use complex link specifications to express the conditions under which a link between two resources can be deemed to exist. However, understanding such complex link specifications is a challenging task for non-expert users of link discovery frameworks. In this paper, we address this drawback by (...)

NELLIE: Never-Ending Linking for Linked Open Data

Authors: Abdullah Fathi Ahmed, Mohamed Ahmed Sherif, Axel-Cyrille Ngonga Ngomo

Knowledge graphs (KGs) that follow the Linked Data principles are created daily. However, there are no holistic models for the Linked Open Data (LOD). Building these models( i.e., engineering a pipeline system) is still a big challenge in order to make the LOD vision comes true. In this paper, we address this challenge by presenting NELLIE (...)

Tutorial: Interactive Adaptive Learning

Authors: Mirko Bunse, Georg Krempl, Alaa Tharwat, Amal Saadallah

We summarize the contents of the tutorial we present as a part of the 7th Interactive Adaptive Learning workshop. This workshop is co-located with the ECML-PKDD conference, where it takes place on September 22nd, 2023 in Turin, Italy.

Beyond the Bias: Unveiling the Quality of Implicit Causality Prompt Continuations in Language Models

Authors: Judith Sieker, Oliver Bott, Torgrim Solstad, Sina Zarrieß

Recent studies have used human continuations of Implicit Causality (IC) prompts collected in linguistic experiments to evaluate discourse understanding in large language models (LLMs), focusing on the well-known IC coreference bias in the LLMs’ predictions of the next word following the prompt. In this study, we investigate how continuations of IC prompts can be used to evaluate the text generation capabilities of LLMs in a linguistically controlled setting. We conduct an experiment using two open-source GPT-based models (...)

Neuro-Symbolic Class Expression Learning

Authors: Caglar Demir, Axel-Cyrille Ngonga Ngomo

Models computed using deep learning have been effectively applied to tackle various problems in many disciplines. Yet, the predictions of these models are often at most post-hoc and locally explainable. In contrast, class expressions in description logics are ante-hoc and globally explainable. Although state-of-the-art symbolic machine learning approaches are being successfully applied to learn class expressions, their application at large scale has been hindered by their impractical runtimes. Arguably, the reliance on myopic heuristic functions contributes to this limitation. We propose (...)

Native Execution of GraphQL Queries over RDF Graphs Using Multi-way Joins

Authors: Nikolaos Karalis, Alexander Bigerl, Axel-Cyrille Ngonga Ngomo

The query language GraphQL has gained significant traction in recent years. In particular, it has recently gained the attention of the semantic web and graph database communities and is now often used as a means to query knowledge graphs. Most of the storage solutions that support GraphQL rely on a translation layer to map the said language to another query language that they support natively, for example SPARQL. (...)

Clifford Embeddings – A Generalized Approach for Embedding in Normed Algebras

Authors: Caglar Demir, Axel-Cyrille Ngonga Ngomo

A growing number of knowledge graph embedding models exploit the characteristics of division algebras (e.g., R, C, H, and O) to learn embeddings. Yet, recent empirical results suggest that the suitability of algebras is contingent upon the knowledge graph being embedded. In this work, we tackle the challenge of selecting the algebra within which a given knowledge graph should be embedded by exploiting the fact that Clifford algebras (...)

COBALT: A Content-Based Similarity Approach for Link Discovery over Geospatial Knowledge Graphs

Authors: Alexander Becker, Abdullah Ahmed, Mohamed Ahmed Sherif, Axel-Cyrille Ngonga Ngomo

Data integration and applications across knowledge graphs (KGs) rely heavily on the discovery of links between resources within these KGs. Geospatial link discovery algorithms have to deal with millions of point sets containing billions of points. (...)

LitCQD: Multi-Hop Reasoning in Incomplete Knowledge Graphs with Numeric Literals

Authors: Caglar Demir, Michel Wiebesiek, Renzhong Lu, Axel-Cyrille Ngonga Ngomo, Stefan Heindorf

Most real-world knowledge graphs, including Wikidata, DBpedia, and Yago are incomplete. Answering queries on such incomplete graphs is an important, but challenging problem. Recently, a number of approaches, including complex query decomposition (CQD), have been proposed to answer complex, multi-hop queries with conjunctions and disjunctions on such graphs. However, these approaches only consider graphs consisting of entities and relations, neglecting literal values. In this paper, we propose LitCQD (...)

Neural Class Expression Synthesis in ALCHIQ(D)

Authors: N’Dah Jean Kouagou, Stefan Heindorf, Caglar Demir, Axel-Cyrille Ngonga Ngomo

Class expression learning in description logics has long been regarded as an iterative search problem in an infinite conceptual space. Each iteration of the search process invokes a reasoner and a heuristic function. The reasoner finds the instances of the current expression, and the heuristic function computes the information gain and decides on the next step to be taken. As the size of the background knowledge base grows, search-based approaches for class expression learning become prohibitively slow. Current neural class expression synthesis (NCES) approaches investigate the use of neural networks for class expression learning in the attributive language (...)

A Topic Model for the Data Web

Authors: Michael Röder, Denis Kuchelev, Axel Ngonga

The usage of knowledge graphs in industry and at Web scale has increased steadily within recent years. However, the decentralized approach to data creation which underpins the popularity of knowledge graphs also comes with significant challenges. In particular, gaining an overview of the topics covered by existing datasets manually becomes a gargantuan if not impossible feat. Several dataset catalogs (...)

Adaptive local Principal Component Analysis improves the clustering of high-dimensional data

Authors: Nico Migenda, Ralf Möller, Wolfram Schenck

In local Principal Component Analysis (PCA), a distribution is approximated by multiple units, each representing a local region by a hyper-ellipsoid obtained through PCA. We present an extension for local PCA which adaptively adjusts both the learning rate of each unit and the potential function which guides the competition between the local units. Our local PCA method is an online neural network method where (...)

Predicting grounding state for adaptive explanation generation in analogical problem-solving

Authors: Lina Mavrina, Stefan Kopp

This paper’s main contribution is a Bayesian hierarchical grounding state prediction model implemented in an adaptive explainer agent assisting users with analogical problem-solving. This model lets the agent adapt dialogue moves regarding previously unmentioned domain entities that are similar to the ones already explained when they are instances of the same generalised schema in different domains. Learning such schemata facilitates knowledge transfer between domains and plays an important role in analogical reasoning (...)

Active Learning for Regression Problems with Ensemble Methods

Authors: Bjarne Jaster, Martin Kohlhase

Traditional machine learning paradigms depend on the availability of labeled data, a luxury that is not often the reality in real-world scenarios. In domains such as industry, healthcare, autonomous systems and finances a massive amount of unlabeled data is produced every day. As the demand for accurate and robust models to deal with this data grows, the inefficiency and the cost of manual labeling motivates the research field active learning (...)

Robust Training with Adversarial Examples on Industrial Data

Authors: Julian Knaup, Christoph-Alexander Holst, Volker Lohweg

In an era where deep learning models are increasingly deployed in safety-critical domains, ensuring their reliability is paramount. The emergence of adversarial examples, which can lead to severe model misbehavior, underscores this need for robustness. Adversarial training, a technique aimed at fortifying models against such threats, is of particular interest. This paper presents an approach tailored to adversarial training on tabular data within industrial environments.

Being ignored is not the only possible form of social exclusion in human-agent interaction

Authors: Clarissa Sabrina Arlinghaus, Günter W. Maier

In a world where humans and technical agents (e.g., robots, AI) work collaboratively, processes of social inclusion and exclusion in human-agent interaction (HAI) gain importance. However, the current focus of social exclusion in HAI is too narrowminded and neglects many forms of social exclusion (e.g., averted eye gazes, microaggressions, hurtful laughter). To change this, the effects of different types of social exclusion will be explored in a series of experiments against the background of William's need-threat mode (...)

Social exclusion in personnel selection – The risk of discriminating AI biases

Authors: Clarissa Sabrina Arlinghaus, Günter W. Maier

Work plays a central role in the life of adults as it opens up access to a wide range of valuable resources (e.g., financial security, time structure, social contacts). Thereby work contributes to the social inclusion of people in most societies. Therefore, personnel selection processes carry a high level of social responsibility. Nowadays, artificial intelligence (AI) is widely used in human resources (HR), but the unreflected use of AI in recruitment can lead to the exclusion of vulnerable groups. (...)

Adaptive Koopman-Based Models for Holistic Controller and Observer Design

Authors: Annika Junker, Keno Pape, Julia Timmermann, Ansgar Trächtler

We present a method to obtain a data-driven Koopman operator-based model that adapts itself during operation and can be straightforwardly used for the controller and observer design. The adaptive model is able to accurately describe different state-space regions and additionally consider unpredictable system changes that occur during operation. Furthermore, we show that this adaptive model is applicable to state-space control, which requires complete knowledge of the state vector. (...)

When Your Language Model Cannot Even Do Determiners Right: Probing for Anti-Presuppositions and the Maximize Presupposition! Principle

Authors: Judith Sieker, Sina Zarrieß

The increasing interest in probing the linguistic capabilities of large language models (LLMs) has long reached the area of semantics and pragmatics, including the phenomenon of presuppositions. In this study, we investigate a phenomenon that, however, has not yet been investigated, i.e., the phenomenon of anti-presupposition and the principle that accounts for it, the Maximize Presupposition! principle (MP!). (...)

TEMPORALFC: A Temporal Fact Checking approach over Knowledge Graphs

Authors: Umair Qudus, Michael Röder, Sabrina Kirrane, Axel-Cyrille Ngonga Ngomo

Verifying assertions is an essential part of creating and maintaining knowledge graphs. Most often, this task cannot be carried out manually due to the sheer size of modern knowledge graphs. Hence, automatic fact-checking approaches have been proposed over the last decade. These approaches aim to compute automatically whether a given assertion is correct or incorrect. (...)

Layered Neural Networks with GELU Activation, a Statistical Mechanics Analysis

Authors: Frederieke Richert, Michiel Straat, Elisa Oostwal, Michael Biehl

Understanding the influence of activation functions on the learning behaviour of neural networks is of great practical interest. The GELU, being similar to swish and ReLU, is analysed for soft committee machines in the statistical physics framework of off-line learning. We find phase transitions with respect to the relative training set size, which are always continuous. This result rules out the hypothesis that convexity is necessary for continuous phase transitions. Moreover, we show that even a small contribution of a sigmoidal function like erf in combination with GELU leads to a discontinuous transition.

Towards Detecting Lexical Change of Hate Speech in Historical Data

Authors: Sanne Hoeken, Sophie Spliethoff, Silke Schwandt, Sina Zarrieß, Özge Alacam

The investigation of lexical change has predominantly focused on generic language evolution, not suited for detecting shifts in a particular domain, such as hate speech. Our study introduces the task of identifying changes in lexical semantics related to hate speech within historical texts. We present an interdisciplinary approach that brings together NLP and History, yielding a pilot dataset comprising 16th-century Early Modern English religious writings during the Protestant Reformation. We provide annotations for both semantic shifts and hatefulness on this data and, thereby, combine the tasks of Lexical Semantic Change Detection and Hate Speech Detection. Our framework and resulting dataset facilitate the evaluation of our applied methods, advancing the analysis of hate speech evolution.

Methodological Insights in Detecting Subtle Semantic Shifts with Contextualized and Static Language Models

Authors: Sanne Hoeken, Özge Alacam, Antske Fokkens, Pia Sommerauer

In this paper, we investigate automatic detection of subtle semantic shifts between social communities of different political convictions in Dutch and English. We perform a methodological study comparing methods using static and contextualized language models. We investigate the impact of specializing contextualized models through fine-tuning on target corpora, word sense disambiguation and sentiment. We furthermore propose a new approach using masked token prediction, that relies on behavioral information, specifically the most probable substitutions, instead of geometrical comparison of representations. Our results show (...)

Supporting wound infection diagnosis: advancements and challenges with electronic noses

Authors: Julius Wörner, Maurice Moelleken, Joachim Dissemon, Miriam Pein-Hackelbusch

Wound infections are a major problem worldwide, both for the healthcare system and for patients affected. Currently available diagnostic methods to determine the responsible germs are time-consuming and costly. Wound infections are mostly caused by various bacteria, which in turn produce volatile organic compounds. From clinical experience, we know that depending on the bacteria involved, a specific odor impression can be expected. For this reason, we hypothesized that electronic noses, (...)

Key Indicators for the Discrimination of Wines by Electronic Noses

Authors: Julius Wörner, Helene Dörksen, Miriam Pein-Hackelbusch

In the food industry, and especially in wines as products thereof, ethanol and sulfur dioxide play an equally important role. Both substances are important wine quality characteristics as they influence the taste and odor. As both substances comprise volatile matter, electronic noses should be applicable to discriminate the different qualities of wines. Our study investigates the influence of alcohol and sulfur dioxide on the discrimination ability of wines (especially those of the same grape variety) using two different electronic nose systems. (...)

Gaussian Process Priors for Systems of Linear Partial Differential Equations with Constant Coefficients

Authors: Marc Härkönen, Markus Lange-Hegermann, Bogdan Raiță

Partial differential equations (PDEs) are important tools to model physical systems and including them into machine learning models is an important way of incorporating physical knowledge. Given any system of linear PDEs with constant coefficients, we propose a family of Gaussian process (GP) priors, which we call EPGP, such that all realizations are exact solutions of this system. We apply the Ehrenpreis-Palamodov fundamental principle (...)

Partial observations, coarse graining and equivariance in Koopman operator theory for large-scale dynamical systems

Authors: Sebastian Peitz, Hans Harder, Feliks Nüske, Friedrich Philipp, Manuel Schaller, Karl Worthmann

The Koopman operator has become an essential tool for data-driven analysis, prediction and control of complex systems, the main reason being the enormous potential of identifying linear function space representations of nonlinear dynamics from measurements. Until now, the situation where for large-scale systems, we (i) only have access to partial observations (i.e., measurements, as is very common for experimental data) or (ii) deliberately perform coarse graining (for efficiency reasons) has not been treated to its full extent. In this paper, we address the pitfall associated (...)