KI 2023: Advances in Artificial Intelligence: 46th German Conference on AI, Berlin, Germany, September 26–29, 2023, Proceedings (Lecture Notes in Artificial Intelligence)

This book constitutes the refereed proceedings of the 46th German Conference on Artificial Intelligence, KI 2023, which took place in Berlin, Germany, in September 2023.The 14 full and 5 short papers presented were carefully reviewed and selected from 78 submissions. The papers deal with research on theory and applications across all methods and topic areas of AI research. Preface Organization Abstracts of Invited Talks Can We Verify That Neural-Network-Based AIs are Ethically Correct? Chatbots and Large Language Models – How Advances in Generative AI Impact Users, Organizations and Society Near-Miss Explanations to Teach Humans and Machines Contents LESS is More: LEan Computing for Selective Summaries 1 Introduction 2 Preliminaries 2.1 Notations 2.2 Subjective Content Descriptions 2.3 Unsupervised Estimator for SCD Matrices 3 Computing Labels for SCDs 3.1 Relations and SCDs 3.2 Labels to Select from 3.3 Utility of Sentences as Labels 3.4 Algorithm LESS 4 Evaluation 4.1 Dataset 4.2 Approaches Using BERT 4.3 Hardware and Metrics 4.4 Workflow and Implementation 4.5 Results 5 Conclusion References Optimisation of Matrix Production System Reconfiguration with Reinforcement Learning 1 Introduction 2 System Definition 3 Results 4 Discussion and Conclusion References CHA2: CHemistry Aware Convex Hull Autoencoder Towards Inverse Molecular Design 1 Introduction 2 Relevant Prior Works 3 Proposed Method and Experimental Design 4 Results 5 Conclusion and Future Work References Ontology Pre-training for Poison Prediction 1 Introduction 2 Methods 2.1 Step 1: Standard Pre-training 2.2 Step 2: Ontology Pre-training 2.3 Step 3: Toxicity Prediction 3 Results 3.1 Predictive Performance 3.2 Interpretability 4 Discussion 4.1 Significance 4.2 Related Work 4.3 Limitations 5 Conclusion References A Novel Incremental Learning Strategy Based on Synthetic Data Generated from a Random Forest 1 Introduction 2 Related Work: Incremental Learning 3 The Proposed Incrementing Strategy: IGTLGSS 3.1 Nearest Class Mean Forest (NCMF) 3.2 IGT Incrementing Strategy 3.3 IGTLGSS Incrementing Strategy 4 Synthetic Data Generation 4.1 Nearest Class Mean Forest (NCMF) 4.2 Gaussian Mixture Model (GMM) 5 Experiments 5.1 Dataset Description 5.2 Experimental Protocol 5.3 Results and Analysis 6 Conclusion and Perspectives References RECol: Reconstruction Error Columns for Outlier Detection 1 Introduction 2 Related Work 3 Approach 3.1 Possible Design Decisions 4 Evaluation 4.1 Experimental Setup 4.2 Results and Discussion 4.3 Possible Design Decisions 4.4 Parameter Recommendations 5 Conclusion and Future Work References Interactive Link Prediction as a Downstream Task for Foundational GUI Understanding Models 1 Introduction 2 Related Work 3 Methodology 3.1 Datasets 3.2 Heuristic Models 3.3 Supervised Learning Models 3.4 Model Selection 4 Experiments 5 Application: `Suggested Links' Figma Plugin 6 Discussion 7 Conclusion References Harmonizing Feature Attributions Across Deep Learning Architectures: Enhancing Interpretability and Consistency 1 Introduction 2 Related Work 3 Methodology 4 Experiment and Results 5 Conclusion References Lost in Dialogue: A Review and Categorisation of Current Dialogue System Approaches and Technical Solutions 1 Introduction 2 Methods 3 Dialogue Systems 3.1 Objective 3.2 Modality 3.3 Domain 3.4 Architecture 3.5 Model 3.6 Correlations Among Categories 4 Applications and Frameworks 4.1 Intelligent Virtual Assistants (IVAs) 4.2 Commercial Frameworks 4.3 Research Dialogue Systems 4.4 Large Language Models (LLMs) 5 Conclusion and Future Work References Cost-Sensitive Best Subset Selection for Logistic Regression: A Mixed-Integer Conic Optimization Perspective 1 Introduction 2 Related Work and Preliminary Definitions 3 Best Subset Selection 3.1 Cardinality-Constrained Best Subset Selection 3.2 Budget-Constrained Best Subset Selection 4 Data Synthesis for Prognostic Models 5 Experiments 5.1 How Does Our Approach Perform Compared to Other Models? 5.2 Comparing Optimal and Greedy Cardinality-Constrained Selection 5.3 Evaluating Our Cost-Sensitive Selection Algorithm 6 Conclusion References Detecting Floors in Residential Buildings*-1pc 1 Introduction 2 Problem Definition 3 Related Work 4 Methods 4.1 Detection of Façade Elements 4.2 Nearest Neighbours (NN) Approach 4.3 Ransac Approach 5 Evaluation 5.1 Dataset 5.2 Results 5.3 Reasons for False Predictions 6 Discussion, Conclusion and Future Work References A Comparative Study of Video-Based Analysis Using Machine Learning for Polyp Classification 1 Introduction 2 Related Work 3 Methods 3.1 Datasets 3.2 Architectures 3.3 Majority Voting with 2D CNN 4 Results and Discussion 5 Conclusion References Object Anchoring for Autonomous Robots Using the Spatio-Temporal-Semantic Environment Representation SEEREP 1 Introduction 2 Related Work 3 SEEREP Architecture Extension 3.1 Instances 3.2 Data Type Point 4 Application Example 5 Conclusion References Associative Reasoning for Commonsense Knowledge 1 Introduction 2 Related Work 3 Preliminaries and Task Description 4 SInE: A Syntax-Based Selection Strategy 5 Use of Statistical Information for the Selection of Axioms 5.1 Distributional Semantics 5.2 Vector-Based Selection: A Statistical Selection Strategy 6 Evaluation: A Case Study on Commonsense Knowledge 6.1 Functional Remote Association Tasks 6.2 Experimental Results 7 Conclusion and Future Work References Computing Most Likely Scenarios of Qualitative Constraint Networks*-1pc 1 Introduction 2 Preliminaries and Problem Definition 2.1 Qualitative Constraint-Based Reasoning 2.2 Problem Definition 2.3 MAX-QCN Problem 3 Monte Carlo Tree Search for Most Likely Scenarios 4 Experimental Evaluation 4.1 Implementation Details 4.2 Dataset 4.3 Experimental Setup 4.4 Experimental Results 5 Conclusion and Future Work References PapagAI: Automated Feedback for Reflective Essays 1 Introduction 2 Related Work 3 Methods, Components and Performances 4 System Architecture 5 Comparison with GPT-3 6 Discussion and Conclusion References Generating Synthetic Dialogues from Prompts to Improve Task-Oriented Dialogue Systems*-1pc 1 Introduction 2 Motivation 3 Method 4 Results and Discussion 5 Conclusion References Flexible Automation of Quantified Multi-Modal Logics with Interactions*-1pc 1 Introduction 2 Preliminaries 2.1 First-Order Multi-Modal Logic 2.2 Higher-Order Logic 2.3 The TPTP Infrastructure for ATP Systems 3 Axiomatization of Modal Logics 3.1 Representation in TPTP 4 Automation via Translation to HOL 4.1 Correctness 4.2 Implementation of the Embedding 5 Application Example 6 Conclusion References Planning Landmark Based Goal Recognition Revisited: Does Using Initial State Landmarks Make Sense?*-1pc 1 Introduction 2 Background 2.1 Classical Planning 2.2 Goal Recognition 2.3 Planning Landmarks 2.4 Extraction of Planning Landmarks 3 Ignoring Initial State Landmarks in Planning Landmark Based Goal Recognition 3.1 Planning Landmark Based Goal Recognition 3.2 Estimating Goal Probabilities 3.3 Why Using Initial State Landmarks Does Bias Goal Recognition Performance 4 Evaluation 4.1 Experimental Design 4.2 Results 5 Related Work 6 Conclusion References Short Papers Explanation-Aware Backdoors in a Nutshell 1 Introduction 2 Explanation-Aware Backdooring Attacks 3 Conclusion References Socially Optimal Non-discriminatory Restrictions for Continuous-Action Games 1 Introduction 2 Model 3 Experiment: The Cournot Game 4 Summary References Few-Shot Document-Level Relation Extraction References Pseudo-Label Selection Is a Decision Problem 1 Introduction: Pseudo-Labeling 2 Approximatley Bayes-Optimal Pseudo-Label Selection 3 In All Likelihoods: Robust Pseudo-Label Selection References XAI Methods in the Presence of Suppressor Variables: A Theoretical Consideration 1 Introduction 2 Results and Discussion 3 Conclusion References Author Index