Article
Computer Science, Artificial Intelligence
Lun Ai, Stephen H. Muggleton, Celine Hocquette, Mark Gromowski, Ute Schmid
Summary: This paper explores the explanatory effects of machine learned theories in human learning, proposing a framework to identify the harmfulness of machine explanations based on the cognitive window concept. Empirical evidence shows that human performance is significantly improved when aided by a symbolic machine learned theory that satisfies the cognitive window, while performance declines when aided by a theory that fails to satisfy the window.
Article
Computer Science, Artificial Intelligence
Lun Ai, Johannes Langer, Stephen H. Muggleton, Ute Schmid
Summary: The comprehensibility of machine-learned theories has gained attention, particularly in the context of logic programming. Previous studies have shown the potential for improving human comprehension with machine-learned logic rules. However, the presentation of machine-learned explanations in game learning can have both positive and negative effects. In this research, the effects of concept ordering and the presence of machine-learned explanations on human comprehension in sequential problem-solving were examined. The results suggest that the sequential teaching of concepts and the presence of explanations can enhance human comprehension and problem-solving strategies.
Article
Computer Science, Artificial Intelligence
S. Patsantzis, S. H. Muggleton
Summary: The paper introduces a method for Meta-Interpretive Learners to construct a Top program instead of searching for correct hypotheses, showing that the algorithm can construct a correct Top program in polynomial time and improve predictive accuracy in large hypothesis spaces or in the presence of classification noise.
Article
Computer Science, Artificial Intelligence
Rolf Morel, Andrew Cropper
Summary: Scientists form hypotheses and experimentally test them. If a hypothesis fails, they try to explain the failure to eliminate other hypotheses. This study introduces failure explanation techniques for inductive logic programming, where a hypothesis is tested on examples and failure is explained in terms of failing sub-programs. The algorithm based on SLD-tree analysis shows that fine-grained failure analysis reduces hypothesis space exploration and learning times.
Article
Computer Science, Artificial Intelligence
Celine Hocquette, Andrew Cropper
Summary: A magic value in a program is an essential constant symbol without clear explanation. Learning programs with magic values is difficult, so we introduce an inductive logic programming approach to tackle this problem. Our experiments show that our approach outperforms existing methods in terms of accuracy and learning time, can handle magic values from infinite domains, and scale to domains with millions of constant symbols.
Article
Computer Science, Artificial Intelligence
Kun Gao, Hanpin Wang, Yongzhi Cao, Katsumi Inoue
Summary: The paper introduces a novel differentiable inductive logic programming system called D-LFIT, with characteristics including a small number of parameters, the ability to generate logic programs in a curriculum-learning setting, and linear time complexity for the extraction of trained neural networks.
Article
Computer Science, Artificial Intelligence
S. Patsantzis, S. H. Muggleton
Summary: This work shows that second-order metarules for Meta-interpretive learning can be learned automatically, replacing the user-defined metarules. It also presents a new specialization operator to achieve this goal.
Article
Computer Science, Artificial Intelligence
Andrew Cropper, Sebastijan Dumancic, Richard Evans, Stephen H. Muggleton
Summary: Inductive Logic Programming (ILP) is a form of logic-based machine learning that focuses on inducing hypotheses that generalize given training examples and background knowledge. Recent research has highlighted new meta-level search methods, techniques for learning recursive programs, approaches for predicate invention, and the use of different technologies. Current limitations of ILP and directions for future research are also being discussed.
Article
Medicine, General & Internal
Soo Young Kim, Young-Il Kim, Hee Jun Kim, Hojin Chang, Seok-Mo Kim, Yong Sang Lee, Soon-Sun Kwon, Hyunjung Shin, Hang-Seok Chang, Cheong Soo Park
Summary: This study used inductive logic programming to predict recurrence among thyroid cancer patients, identifying three rules that could predict recurrence, with postoperative thyroglobulin level being the most powerful variable. These rules demonstrated high accuracy in predicting recurrence during validation.
Article
Computer Science, Software Engineering
Huaduo Wang, Farhad Shakerin, Gopal Gupta
Summary: FOLD-RM is an automated inductive learning algorithm that generates explainable models and human-friendly explanations for learning default rules for mixed data.
THEORY AND PRACTICE OF LOGIC PROGRAMMING
(2022)
Article
Computer Science, Artificial Intelligence
Ludovico Mitchener, David Tuckey, Matthew Crosby, Alessandra Russo
Summary: In this paper, we present DUA, a neuro-symbolic reinforcement learning framework that combines computer vision, inductive logic programming, and deep reinforcement learning. By integrating these techniques, we address physical cognitive reasoning problems and establish foundations for tackling complex DRL challenges.
Article
Computer Science, Artificial Intelligence
Daniele Meli, Mohan Sridharan, Paolo Fiorini
Summary: The study investigates the use of logic programming in task planning for robot-assisted surgery to improve performance. By learning under event calculus formalism, a systematic approach for learning the specifications of a generic robotic task is proposed, allowing for easy knowledge refinement through iterative learning. The learned axioms demonstrate significant improvement in performance compared to hand-written ones, especially addressing critical issues related to plan computation time for reliable real-time performance during surgery.
Article
Computer Science, Software Engineering
Nitesh Kumar, Ondrej Kuzelka, Luc De Raedt
Summary: This article introduces a method for relational autocompletion using the Distributional Clauses framework for probabilistic logic programming, combined with statistical modeling and rule learning. The empirical results demonstrate the effectiveness of the approach, even in the presence of missing data.
THEORY AND PRACTICE OF LOGIC PROGRAMMING
(2022)
Article
Computer Science, Artificial Intelligence
Ashwin Srinivasan, Michael Bain, A. Baskar
Summary: This paper proposes a method for identifying feedback mechanisms in biological systems by learning kinetic logic, formalized as a labeled transition system and implemented in a modified form of event calculus. The approach allows for specifying system identification and identifying regulatory mechanisms in biological problems by combining induction and abduction techniques.
Article
Computer Science, Artificial Intelligence
Alice Tarzariol, Martin Gebser, Konstantin Schekotihin
Summary: Efficiently excluding symmetric solution candidates is crucial for combinatorial problem-solving. Existing approaches that compute Symmetry Breaking Constraints (SBCs) for specific problem instances may not be applicable to large-scale instances or advanced problem encodings. To overcome these limitations, we propose a model-oriented approach that transforms SBCs of small problem instances into interpretable first-order constraints using Inductive Logic Programming.
Editorial Material
Statistics & Probability
Daniel Baier, Berthold Lausen, Angela Montanari, Ute Schmid
ADVANCES IN DATA ANALYSIS AND CLASSIFICATION
(2020)
Article
Statistics & Probability
Mark Gromowski, Michael Siebers, Ute Schmid
Summary: A general process framework for logic-rule-based classifiers has been developed to facilitate mutual exchange between system and user, allowing users to detail explain system decisions and correct errors. The framework suggests integrating users' corrections into the system's core logic rules through retraining to enhance system performance.
ADVANCES IN DATA ANALYSIS AND CLASSIFICATION
(2020)
Article
Computer Science, Artificial Intelligence
Lun Ai, Stephen H. Muggleton, Celine Hocquette, Mark Gromowski, Ute Schmid
Summary: This paper explores the explanatory effects of machine learned theories in human learning, proposing a framework to identify the harmfulness of machine explanations based on the cognitive window concept. Empirical evidence shows that human performance is significantly improved when aided by a symbolic machine learned theory that satisfies the cognitive window, while performance declines when aided by a theory that fails to satisfy the window.
Article
Computer Science, Artificial Intelligence
Teena Hassan, Dominik Seuss, Johannes Wollenberg, Katharina Weitz, Miriam Kunz, Stefan Lautenbacher, Jens-Uwe Garbas, Ute Schmid
Summary: Pain sensation is crucial for survival and observer reports are important for noncommunicative patients. Automatic pain detection technology can assist human caregivers and improve pain management. Facial expressions are reliable indicators of pain, and computer vision researchers are using this technology to automatically detect pain.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Software Engineering
Joscha Eirich, Jakob Bonart, Dominik Jaeckle, Michael Sedlmair, Ute Schmid, Kai Fischbach, Tobias Schreck, Jurgen Bernard
Summary: IRVINE, a Visual Analytics system, helps detect and understand previously unknown errors in the manufacturing of electrical engines. By leveraging interactive clustering and data labeling techniques, users can quickly analyze engine data and annotate errors. Through field studies, IRVINE has shown significant improvements in manufacturing efficiency.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Article
Computer Science, Artificial Intelligence
Johannes Rabold, Michael Siebers, Ute Schmid
Summary: Recent research has shown a growing interest in human-understandable explanations of machine learning models. Aligning examples with similar counterexamples can enhance concept understanding. Generating explanations by modifying rules to cover near misses instead of original instances is a proposed method.
Article
Clinical Neurology
Stefan Lautenbacher, Teena Hassan, Dominik Seuss, Frederik W. Loy, Jens-Uwe Garbas, Ute Schmid, Miriam Kunz
Summary: This study compared manual and automatic coding of AU for facial expressions of pain. The results showed poor outcomes for automatic AU coding in terms of sensitivity/recall, precision, and F1. Congruency was better for younger faces and pain-indicative AUs.
PAIN RESEARCH & MANAGEMENT
(2022)
Editorial Material
Computer Science, Artificial Intelligence
Ute Schmid, Katharina Rohlfing, Philipp Cimiano
KUNSTLICHE INTELLIGENZ
(2022)
Article
Computer Science, Software Engineering
Joscha Eirich, Dominik Jaeckle, Michael Sedlmair, Christoph Wehner, Ute Schmid, Jurgen Bernard, Tobias Schreck
Summary: We present ManuKnowVis, a design study that contextualizes data from multiple knowledge repositories to enhance data-driven analyses in the manufacturing process of battery modules for electric vehicles. Our study reveals a discrepancy between knowledge providers, who have domain knowledge but struggle with data-driven analyses, and knowledge consumers, who lack domain knowledge but excel in data analyses. ManuKnowVis bridges this gap by enabling collaboration between providers and consumers and facilitating the creation and completion of manufacturing knowledge. The tool incorporates multiple linked views, allowing providers to describe and connect individual entities based on their domain knowledge, and consumers to leverage this enhanced data for more efficient data analyses.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2023)
Article
Computer Science, Artificial Intelligence
Simon Schramm, Christoph Wehner, Ute Schmid
Summary: This article introduces the historical development of understandable artificial intelligence on knowledge graphs, distinguishes the concepts of understandable artificial intelligence and interpretable machine learning, and proposes a new taxonomy. The article also summarizes the current research on understandable artificial intelligence on knowledge graphs and identifies future research directions.
JOURNAL OF WEB SEMANTICS
(2023)
Proceedings Paper
Computer Science, Information Systems
Julio Wissing, Stephan Scheele, Aliya Mohammed, Dorothea Kolossa, Ute Schmid
Summary: Smart sensor systems enable machine learning algorithms to be executed at the data source, which is particularly useful for moving parts or remote areas. However, this increases the power load on the measurement device. The HiMLEdge framework introduces hierarchical machine learning with energy-aware optimization, achieving specialized models. Tests show that the optimized hierarchical model can perform more readings with the same battery life compared to a flat classifier.
ADVANCED RESEARCH IN TECHNOLOGIES, INFORMATION, INNOVATION AND SUSTAINABILITY, ARTIIS 2022, PT II
(2022)
Article
Computer Science, Artificial Intelligence
Sebastian Kiefer, Mareike Hoffmann, Ute Schmid
Summary: Interactive Machine Learning (IML) is becoming increasingly relevant in various application domains. This study proposes a novel framework called Semantic Interactive Learning for document classification, which incorporates constructive and contextual feedback into the learner. By introducing the SemanticPush technique, human conceptual corrections are effectively translated into non-extrapolating training examples, pushing the learner's reasoning towards the desired behavior.
MACHINE LEARNING AND KNOWLEDGE EXTRACTION
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Christoph Wehner, Francis Powlesland, Bashar Altakrouri, Ute Schmid
Summary: This research presents an approach to explaining lane change predictions using Layer-wise Relevance Propagation and communicates the explanations to users through real-time data and an interface.
ADVANCES AND TRENDS IN ARTIFICIAL INTELLIGENCE: THEORY AND PRACTICES IN ARTIFICIAL INTELLIGENCE
(2022)
Proceedings Paper
Automation & Control Systems
Deniz Neufeld, Ute Schmid
Summary: This study focuses on computationally efficient difference metrics of time series and compares two different unsupervised methods for anomaly classification in the domain of hardware systems testing. The research found that Mean Squared Error towards the median in combination with the Modified z-Score is the most robust method for detecting anomalies, especially under concept drift.
2021 26TH IEEE INTERNATIONAL CONFERENCE ON EMERGING TECHNOLOGIES AND FACTORY AUTOMATION (ETFA)
(2021)
Article
Computer Science, Artificial Intelligence
Ute Schmid, Bettina Finzel
KUNSTLICHE INTELLIGENZ
(2020)