TY - CONF
AB - 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 as a translation problem and propose a new family of class expression learning approaches which we dub neural class expression synthesizers. Training examples are ``translated'' into class expressions in a fashion akin to machine translation. Consequently, our synthesizers are not subject to the runtime limitations of search-based approaches. We study three instances of this novel family of approaches based on LSTMs, GRUs, and set transformers, respectively. An evaluation of our approach on four benchmark datasets suggests that it can effectively synthesize high-quality class expressions with respect to the input examples in approximately one second on average. Moreover, a comparison to state-of-the-art approaches suggests that we achieve better F-measures on large datasets. For reproducibility purposes, we provide our implementation as well as pretrained models in our public GitHub repository at https://github.com/dice-group/NeuralClassExpressionSynthesis
AU - KOUAGOU, N'Dah Jean
AU - Heindorf, Stefan
AU - Demir, Caglar
AU - Ngonga Ngomo, Axel-Cyrille
ED - Pesquita, Catia
ED - Jimenez-Ruiz, Ernesto
ED - McCusker, Jamie
ED - Faria, Daniel
ED - Dragoni, Mauro
ED - Dimou, Anastasia
ED - Troncy, Raphael
ED - Hertling, Sven
ID - 33734
KW - Neural network
KW - Concept learning
KW - Description logics
T2 - The Semantic Web - 20th Extended Semantic Web Conference (ESWC 2023)
TI - Neural Class Expression Synthesis
VL - 13870
ER -
TY - GEN
AB - Knowledge bases are widely used for information management on the web,
enabling high-impact applications such as web search, question answering, and
natural language processing. They also serve as the backbone for automatic
decision systems, e.g. for medical diagnostics and credit scoring. As
stakeholders affected by these decisions would like to understand their
situation and verify fair decisions, a number of explanation approaches have
been proposed using concepts in description logics. However, the learned
concepts can become long and difficult to fathom for non-experts, even when
verbalized. Moreover, long concepts do not immediately provide a clear path of
action to change one's situation. Counterfactuals answering the question "How
must feature values be changed to obtain a different classification?" have been
proposed as short, human-friendly explanations for tabular data. In this paper,
we transfer the notion of counterfactuals to description logics and propose the
first algorithm for generating counterfactual explanations in the description
logic $\mathcal{ELH}$. Counterfactual candidates are generated from concepts
and the candidates with fewest feature changes are selected as counterfactuals.
In case of multiple counterfactuals, we rank them according to the likeliness
of their feature combinations. For evaluation, we conduct a user survey to
investigate which of the generated counterfactual candidates are preferred for
explanation by participants. In a second study, we explore possible use cases
for counterfactual explanations.
AU - Sieger, Leonie Nora
AU - Heindorf, Stefan
AU - Blübaum, Lukas
AU - Ngonga Ngomo, Axel-Cyrille
ID - 37937
T2 - arXiv:2301.05109
TI - Counterfactual Explanations for Concepts in ELH
ER -
TY - GEN
AB - Label noise poses an important challenge in machine learning, especially in
deep learning, in which large models with high expressive power dominate the
field. Models of that kind are prone to memorizing incorrect labels, thereby
harming generalization performance. Many methods have been proposed to address
this problem, including robust loss functions and more complex label correction
approaches. Robust loss functions are appealing due to their simplicity, but
typically lack flexibility, while label correction usually adds substantial
complexity to the training setup. In this paper, we suggest to address the
shortcomings of both methodologies by "ambiguating" the target information,
adding additional, complementary candidate labels in case the learner is not
sufficiently convinced of the observed training label. More precisely, we
leverage the framework of so-called superset learning to construct set-valued
targets based on a confidence threshold, which deliver imprecise yet more
reliable beliefs about the ground-truth, effectively helping the learner to
suppress the memorization effect. In an extensive empirical evaluation, our
method demonstrates favorable learning behavior on synthetic and real-world
noise, confirming the effectiveness in detecting and correcting erroneous
training labels.
AU - Lienen, Julian
AU - Hüllermeier, Eyke
ID - 45244
T2 - arXiv:2305.13764
TI - Mitigating Label Noise through Data Ambiguation
ER -
TY - JOUR
AB - This article presents the potential-dependent adsorption of two proteins, bovine serum albumin (BSA) and lysozyme (LYZ), on Ti6Al4V alloy at pH 7.4 and 37 °C. The adsorption process was studied on an electropolished alloy under cathodic and anodic overpotentials, compared to the open circuit potential (OCP). To analyze the adsorption process, various complementary interface analytical techniques were employed, including PM-IRRAS (polarization-modulation infrared reflection-absorption spectroscopy), AFM (atomic force microscopy), XPS (X-ray photoelectron spectroscopy), and E-QCM (electrochemical quartz crystal microbalance) measurements. The polarization experiments were conducted within a potential range where charging of the electric double layer dominates, and Faradaic currents can be disregarded. The findings highlight the significant influence of the interfacial charge distribution on the adsorption of BSA and LYZ onto the alloy surface. Furthermore, electrochemical analysis of the protein layers formed under applied overpotentials demonstrated improved corrosion protection properties. These studies provide valuable insights into protein adsorption on titanium alloys under physiological conditions, characterized by varying potentials of the passive alloy.
AU - Duderija, Belma
AU - González-Orive, Alejandro
AU - Ebbert, Christoph
AU - Neßlinger, Vanessa
AU - Keller, Adrian
AU - Grundmeier, Guido
ID - 45828
IS - 13
JF - Molecules
KW - Chemistry (miscellaneous)
KW - Analytical Chemistry
KW - Organic Chemistry
KW - Physical and Theoretical Chemistry
KW - Molecular Medicine
KW - Drug Discovery
KW - Pharmaceutical Science
SN - 1420-3049
TI - Electrode Potential-Dependent Studies of Protein Adsorption on Ti6Al4V Alloy
VL - 28
ER -
TY - CHAP
AU - Keller, Adrian
AU - Grundmeier, Guido
ID - 45829
SN - 9780124095472
T2 - Reference Module in Chemistry, Molecular Sciences and Chemical Engineering
TI - High-speed AFM studies of macromolecular dynamics at solid/liquid interfaces
ER -
TY - CHAP
AU - Kostan, Anastassija
ID - 45833
SN - 9783828877368
T2 - Bedeutung und Implikationen epistemischer Ungerechtigkeit
TI - Die epistemische Gewalt KI-basierter Gesichtserkennung. Wie ein codierter Blick neue Formen der technologisierten Subalternität erschafft
ER -
TY - THES
AB - Reading between the lines has so far been reserved for humans. The present dissertation addresses this research gap using machine learning methods.
Implicit expressions are not comprehensible by computers and cannot be localized in the text. However, many texts arise on interpersonal topics that, unlike commercial evaluation texts, often imply information only by means of longer phrases. Examples are the kindness and the attentiveness of a doctor, which are only paraphrased (“he didn’t even look me in the eye”). The analysis of such data, especially the identification and localization of implicit statements, is a research gap (1). This work uses so-called Aspect-based Sentiment Analysis as a method for this purpose. It remains open how the aspect categories to be extracted can be discovered and thematically delineated based on the data (2). Furthermore, it is not yet explored how a collection of tools should look like, with which implicit phrases can be identified and thus made explicit
(3). Last, it is an open question how to correlate the identified phrases from the text data with other data, including the investigation of the relationship between quantitative scores (e.g., school grades) and the thematically related text (4). Based on these research gaps, the research question is posed as follows: Using text mining methods, how can implicit rating content be properly interpreted and thus made explicit before it is automatically categorized and quantified?
The uniqueness of this dissertation is based on the automated recognition of implicit linguistic statements alongside explicit statements. These are identified in unstructured text data so that features expressed only in the text can later be compared across data sources, even though they were not included in rating categories such as stars or school grades. German-language physician ratings from websites in three countries serve as the sample domain. The solution approach consists of data creation, a pipeline for text processing and analyses based on this. In the data creation, aspect classes are identified and delineated across platforms and marked in text data. This results in six datasets with over 70,000 annotated sentences and detailed guidelines. The models that were created based on the training data extract and categorize the aspects. In addition, the sentiment polarity and the evaluation weight, i. e., the importance of each phrase, are determined. The models, which are combined in a pipeline, are used in a prototype in the form of a web application. The analyses built on the pipeline quantify the rating contents by linking the obtained information with further data, thus allowing new insights.
As a result, a toolbox is provided to identify quantifiable rating content and categories using text mining for a sample domain. This is used to evaluate the approach, which in principle can also be adapted to any other domain.
AU - Kersting, Joschka
ID - 44323
TI - Identifizierung quantifizierbarer Bewertungsinhalte und -kategorien mittels Text Mining
ER -
TY - JOUR
AB - The aim of the present study is to prove the construct validity of the German versions of the Feeling Scale (FS) and the Felt Arousal Scale (FAS) for a progressive muscle relaxation (PMR) exercise. A total of 228 sport science students conducted the PMR exercise for 45 min and completed the FS, the FAS, and the Self-Assessment Manikin (SAM) in a pre-test–post-test design. A significant decrease in arousal (t(227) = 8.296, p < 0.001) and a significant increase in pleasure (t(227) = 4.748, p < 0.001) were observed. For convergent validity, the correlations between the FS and the subscale SAM-P for the valence dimension (r = 0.67, p < 0.001) and between the FAS and the subscale SAM-A for the arousal dimension (r = 0.31, p < 0.001) were significant. For discriminant validity, the correlations between different constructs (FS and SAM-A, FAS and SAM-P) were not significant, whereas the discriminant analysis between the FS and the FAS revealed a negative significant correlation (r = −0.15, p < 0.001). Together, the pattern of results confirms the use of the German versions of the FS and the FAS to measure the affective response for a PMR exercise.
AU - Thorenz, Kristin
AU - Berwinkel, Andre
AU - Weigelt, Matthias
ID - 45857
IS - 7
JF - Behavioral Sciences
KW - Behavioral Neuroscience
KW - General Psychology
KW - Genetics
KW - Development
KW - Ecology
KW - Evolution
KW - Behavior and Systematics
SN - 2076-328X
TI - A Validation Study for the German Versions of the Feeling Scale and the Felt Arousal Scale for a Progressive Muscle Relaxation Exercise
VL - 13
ER -
TY - JOUR
AU - Thorenz, Kristin
AU - Berwinkel, Andre
AU - Weigelt, Matthias
ID - 45856
IS - 06
JF - Psychology
KW - General Earth and Planetary Sciences
KW - General Environmental Science
SN - 2152-7180
TI - A Validation Study of the German Versions of the Feeling Scale and the Felt Arousal Scale for a Passive Relaxation Technique (Autogenic Training)
VL - 14
ER -
TY - JOUR
AU - Topalović, Elvira
AU - Blachut, Alisa
ID - 45861
JF - Der Deutschunterricht
TI - Grammatische Modelle. Einführung in das Themenheft
VL - 3
ER -