Discovery Science: 26th International Conference, DS 2023, Porto, Portugal, October 9–11, 2023, Proceedings (Lecture Notes in Artificial Intelligence)

Preface
Organization
Plenary Talks
Time: The Next Frontier in Discovery Science
Computational Creativity: From Autonomous Generation to Co-creation
Lifelong Anomaly Detection
Contents
Machine Learning Methods and Applications
Ensembles of Classifiers and Quantifiers with Data Fusion for Quantification Learning
	1 Introduction
	2 Background
	3 Related Work
		3.1 Ensembles
	4 Proposed Approaches
	5 Experimental Setup
	6 Results and Discussion
	7 Analysis of the Data Fusion Operators
	8 Conclusions
	References
Exploring the Intricacies of Neural Network Optimization
	1 Introduction
	2 Background
		2.1 Automated Hyperparameter Optimization
		2.2 Hyperparameter Importance
	3 Experiments
		3.1 Dataset Description
		3.2 Baseline Models
		3.3 Hyperparameter Search Space
		3.4 Performance Metrics
		3.5 Individual Hyperparameter Testing
	4 Results
		4.1 General Importance
		4.2 Importance by Dataset Type
	5 Discussion
	6 Conclusion
	A Hyperparameter Importance per Dataset
	B Baseline Models Architecture
	References
Exploring the Reduction of Configuration Spaces of Workflows
	1 Introduction
	2 Relation to Other Work
	3 Reducing the Configuration Space of Workflows
		3.1 Variants of the Reduction Method Considered
		3.2 Using a Given Configuration of Workflows for Recommendation
		3.3 Experiments
	4 Discussion, Future Work and Conclusions
		4.1 Discussion and Future Work
		4.2 Conclusions
	References
iSOUP-SymRF: Symbolic Feature Ranking with Random Forests in Online Multi-target Regression
	1 Introduction
	2 Related Work
	3 Symbolic Feature Ranking with Random Forests
	4 Experimental Setup
		4.1 Datasets
		4.2 Experiment: Parameter Stability
		4.3 Experiment: Ranking Utility
	5 Results
		5.1 Parameter Stability
		5.2 Ranking Utility
	6 Conclusions and Further Work
	References
Knowledge-Guided Additive Modeling for Supervised Regression
	1 Introduction
	2 Problem Statement
	3 Related Work
	4 Methods
		4.1 Sequential Training of hk and ha
		4.2 Alternate Training of hk and ha
		4.3 Partial Dependence-Based Training of hk and ha
	5 Experiments
		5.1 Friedman Problem (A2 and A3 Satisfied)
		5.2 Correlated Input Features (A3 Not Satisfied)
		5.3 Overlapping Additive Structure (A2 and A3 Not Satisfied)
		5.4 Real Regression Problems
	6 Conclusion
	A Optimal Model Under A2 and A3
	B Model Architectures
	References
Natural Language Processing and Social Media Analysis
Audience Prediction for Game Streaming Channels Based on Vectorization of User Comments
	1 Introduction
	2 Related Work
		2.1 Video Streaming
		2.2 Link Prediction
	3 Proposed Method
		3.1 Bipartite Graph and Feature Vector
		3.2 Prediction Model
		3.3 Learning Algorithm
	4 Experimental Settings
		4.1 Dataset
		4.2 Comparison Method
	5 Experimental Results
		5.1 Evaluation of Channel Feature Vectors
		5.2 Prediction Accuracy Comparison
	6 Conclusion
	References
From Tweets to Stance: An Unsupervised Framework for User Stance Detection on Twitter
	1 Introduction
	2 Related Work
	3 Task Definition
		3.1 Data Collection
	4 Methodology
		4.1 Topic Filtering
		4.2 Agreement Detector
		4.3 Experiment Settings
	5 Results and Discussion
		5.1 RQ1: What are the Performances and Insights of T2S?
		5.2 RQ2: Can T2S Generalize over Diverse Political Contexts?
		5.3 Potential and Limitations
	6 Conclusions
	References
GLORIA: A Graph Convolutional Network-Based Approach for Review Spam Detection
	1 Introduction
	2 Related Work
	3 The Proposed Method
	4 Experimental Setup
		4.1 Data
		4.2 Implementation Details
		4.3 Results and Discussion
	5 Conclusion
	References
Unmasking COVID-19 False Information on Twitter: A Topic-Based Approach with BERT
	1 Introduction
	2 Related Work
		2.1 False Information Detection
		2.2 Topic Detection
	3 Proposed Approach
		3.1 Fine-Tuning of the False Information Detection Model
		3.2 Topic Detection
		3.3 Topic Annotation
	4 Experimental Results
		4.1 Model Selection for False Information Detection
		4.2 COVID-Related Detected Topics
		4.3 Topic-Oriented False Information Detected in COVID Discussions
	5 Conclusion
	References
Unsupervised Key-Phrase Extraction from Long Texts with Multilingual Sentence Transformers
	1 Introduction
	2 The Proposed Approaches
		2.1 Text Representations from a Longformer Model Built from a Multilingual Sentence-Transformer
		2.2 LMEmbedRank
		2.3 LMMaskRank
		2.4 Combining Both Ranking Approaches
	3 Experimental Evaluation
		3.1 Metrics and Datasets
		3.2 Experimental Results over the Different Datasets
		3.3 Results for Ablation Experiments
		3.4 Comparison to Previous Methods
	4 Conclusions and Future Work
	References
Interpretability and Explainability in AI
Counterfactuals Explanations for Outliers via Subspaces Density Contrastive Loss
	1 Introduction
	2 Related Works
	3 The Proposed Technique
		3.1 Architecture Pipeline Description
		3.2 Explanation Computation
	4 Experiments
		4.1 Employed Metrics
		4.2 Parameters Tuning
		4.3 Comparison with ATOM and COIN
	5 Conclusions
	References
Explainable Spatio-Temporal Graph Modeling
	1 Introduction
	2 Background
		2.1 XAI Methods
		2.2 Forecasting Methods for Sensor Networks
	3 Method
		3.1 Spatio-Temporal Forecasting Setting
		3.2 Explainability Problem Definition
		3.3 Multi-level Explanation Process
	4 Experiments
		4.1 Datasets
		4.2 Experimental Setup
		4.3 Results and Discussion
	5 Conclusion
	References
Probabilistic Scoring Lists for Interpretable Machine Learning
	1 Introduction
	2 Related Work
	3 From Scoring Systems to Probabilistic Scoring Lists
	4 Learning Probabilistic Scoring Lists
		4.1 A Greedy Learning Algorithm
		4.2 Probability Estimation
		4.3 Beyond Probabilities: Capturing Epistemic Uncertainty
	5 A Case Study in Medical Decision Making
		5.1 Coronary Heart Disease Data
		5.2 Expected Entropy Minimisation
		5.3 Expected Loss Minimisation
	6 Summary and Conclusion
	References
Refining Temporal Visualizations Using the Directional Coherence Loss
	1 Introduction
	2 Related Work
	3 Methods
		3.1 t-SNE
		3.2 Directional Coherence Loss (DCL)
	4 Results and Discussion
		4.1 Toy Example
		4.2 COVID-19 Pandemic in Slovenia
		4.3 Hyperparameter and Evaluation Considerations
	5 Conclusion
	References
Semantic Enrichment of Explanations of AI Models for Healthcare
	1 Introduction
	2 Related Work
	3 Problem Statement
	4 Methodology
		4.1 SNOMED-CT Relationships Extraction
		4.2 Information Extraction from Clinical Notes
	5 Experiments and Results
		5.1 Dataset
		5.2 Implementation Details
		5.3 Human Validated Experiment
	6 Conclusion and Future Work
	References
Text to Time Series Representations: Towards Interpretable Predictive Models
	1 Introduction
	2 Related Works
	3 Setting the Stage
	4 Text to Time Series Conversion
	5 Experiments
	6 Conclusion
	References
Data Analysis and Optimization
Enhancing Intra-modal Similarity in a Cross-Modal Triplet Loss
	1 Introduction
	2 Related Work
	3 Preliminaries
	4 Proposed Methods
		4.1 Motivation
		4.2 Full Hard Negative Method (F-HN):
		4.3 Intra-modal Margin Hard Negative Control Method (M-HN)
	5 Experimental Setting
		5.1 Datasets
		5.2 Models and Implementation Details.
		5.3 Evaluation
	6 Results and Discussion
		6.1 Cross-Modal Retrieval Performance
		6.2 Intra-modal Retrieval Performance
		6.3 Novelty Results
	7 Conclusions and Future Work
	References
Exploring the Potential of Optimal Active Learning via a Non-myopic Oracle Policy
	1 Introduction
	2 Problem Setting
	3 Related Work
	4 Non-myopic Oracle Policy
	5 Experiments
		5.1 Experimental Setup
		5.2 Ablation Studies
		5.3 Comparison with SA Search and Selection Strategies
	6 Conclusion
	References
Extrapolation is Not the Same as Interpolation
	1 Introduction
	2 Method
		2.1 Datasets and Data Pre-processing
		2.2 Formulation of Baseline Approach
		2.3 Formulation of Pairwise Approach
		2.4 Extrapolation Strategy
		2.5 Machine Learning Methods
		2.6 Extrapolation Metrics
	3 Results and Discussion
		3.1 The Pairwise Approach Extrapolates Better
		3.2 The Extrapolation Strategy Improves Extrapolation
		3.3 Discussion
	4 Conclusion
	References
Gene Interactions in Survival Data Analysis: A Data-Driven Approach Using Restricted Mean Survival Time and Literature Mining
	1 Introduction
	2 Methods
		2.1 Data
		2.2 Summary Measure of Survival: Restricted Mean Survival Time
		2.3 Interaction Scoring
		2.4 Interaction Types
		2.5 Discovering False Positives with Permutation Test
		2.6 Literature Mining
	3 Results
		3.1 Analysis Reveals Potential Interactions
		3.2 Cross-Referencing with Established Interaction Networks
		3.3 Case Study: RHOA-CD44 Competing Interaction
	4 Discussion
	5 Conclusions
	References
Joining Imputation and Active Feature Acquisition for Cost Saving on Data Streams with Missing Features
	1 Introduction
	2 Related Works
	3 Methodology
		3.1 Supervised Merit Ranking Active Feature Acquisition Framework
		3.2 Incremental Percentile Filter (IPF)
		3.3 Defining Budget
		3.4 Feature Pair Imputer (FPI)
		3.5 Feature Pair Imputer Threshold Skip (FPITS)
	4 Evaluation
		4.1 Experiment Setup
		4.2 FPI Performance
		4.3 FPITS Behavior
		4.4 Budget Comparison at Similar Performance
	5 Conclusion
	References
Fairness, Privacy and Security in AI
EXPHLOT: EXplainable Privacy Assessment for Human LOcation Trajectories
	1 Introduction
	2 Related Works
		2.1 Explainability
	3 Background
		3.1 Privacy Risk Assessment Framework
		3.2 EXPERT
	4 EXPHLOT
		4.1 EXPHLOT Predictive Model
		4.2 Exphlot Risk Explanation Module
		4.3 Exphlot Risk and Explanation Visualization Module
	5 Experiments
		5.1 Exphlot Privacy Risk Prediction Module
		5.2 Mobility Privacy Risk Explanation
	6 Conclusion
	References
Fairness-Aware Mixture of Experts with Interpretability Budgets
	1 Introduction
	2 Related Work
	3 Fair Mixture of Experts
		3.1 FairMOE Components
	4 Experimental Evaluation
		4.1 Data
		4.2 Algorithms
		4.3 Evaluation Metrics
		4.4 Results
	5 Discussion
	6 Conclusion
	References
GenFair: A Genetic Fairness-Enhancing Data Generation Framework
	1 Introduction
	2 Related Work
	3 Background and Problem Statement
		3.1 Preferential Sampling
		3.2 Genetic Algorithms
	4 GenFair
	5 Experiments
		5.1 Evaluation Metrics
		5.2 Fairness and Performance Evaluation
		5.3 Plausibility Evaluation
	6 Conclusions
	References
Privacy-Preserving Learning of Random Forests Without Revealing the Trees
	1 Introduction
	2 Related Work
	3 Preliminaries
		3.1 Secure Multi-party Computation
		3.2 Arithmetic Secret Shares
	4 Method
		4.1 Privacy-Preserving Determination of Maximum Value
		4.2 Privacy-Preserving Decision Tree Learning
		4.3 Training and Inference for Privacy-Preserving Random Forests
	5 Experiments
	6 Conclusion and Future Work
	References
Unlearning Spurious Correlations in Chest X-Ray Classification
	1 Introduction
	2 Related Work
		2.1 Chest X-Ray Classification
		2.2 eXplanation Based Learning
	3 eXemplary eXplanation Based Learning
	4 Experiments
		4.1 Data Collection and Preparation
		4.2 Model Training
	5 Results
	6 Conclusion
	References
Control and Spatio-Temporal Modeling
Explaining the Chronological Attribution of Greek Papyri Images
	1 Introduction
		1.1 Background
		1.2 The Contributions of This Work
	2 Related Work
	3 Data
		3.1 The Nature of the Papyri
		3.2 Digitised Papyri
		3.3 Our New Dataset
	4 Method
		4.1 fCNN
		4.2 The Baseline
	5 Experiments
		5.1 Experimental Details
	6 Assessing Data Sources Limitations
	7 Error Analysis
	8 Dates in Doubt: A Computational Estimate
	9 Conclusions
	References
Leveraging the Spatiotemporal Analysis of Meisho-e Landscapes
	1 Introduction
		1.1 Research Aims
	2 Related Work
	3 The Corpus of Digitised Meisho-e Prints
	4 Methods
	5 Experiments
	6 Empirical Analysis
		6.1 Inscription Text Restoration
		6.2 Data Augmentation with OCR
		6.3 Geolocating Recognised Place-Name Entities
		6.4 Spatiotemporal Analysis
	7 Conclusion
	References
Predictive Inference Model of the Physical Environment that Emulates Predictive Coding
	1 Introduction
	2 Related Work
	3 PredNet
	4 Variational Temporal Abstraction (VTA)
	5 Proposed Model
		5.1 Mechanism of the Change Point Prediction Model
	6 Experiment
		6.1 Change Point Extraction in Predictive Inference
		6.2 Text Generation of Prediction Results
	7 Conclusions
	References
Transferring a Learned Qualitative Cart-Pole Control Model to Uneven Terrains
	1 Introduction
	2 The Cart-Pole System on Uneven Terrain
	3 Learning a Qualitative Controller for Cart-Pole on Flat Surface
		3.1 Learning a Qualitative Model
		3.2 Finding a Qualitative Plan
	4 Reactive Execution
	5 Transferring the Qualitative Controller from Flat to Uneven Terrains
	6 Conclusions
	References
Which Way to Go - Finding Frequent Trajectories Through Clustering
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Definitions
		3.2 Trajectories Simplification with Ramer-Douglas-Peucker (RDP)
		3.3 Clustering Algorithm and Similarity Measures
	4 Experiments and Results
		4.1 Datasets
		4.2 Data Filtering and Preprocessing
		4.3 Experimental Setup
		4.4 Evaluation
		4.5 Discussion
	5 Conclusions and Future Work
	References
Graph Theory and Network Analysis
Boosting-Based Construction of BDDs for Linear Threshold Functions and Its Application to Verification of Neural Networks
	1 Introduction
	2 Preliminaries
		2.1 Binary Neural Network
		2.2 Definition of BDD, OBDD and ABDD
		2.3 Instance-Based Robustness (IR), Model-Based Robustness (MR) and Sample-Based Robustness (SR)
		2.4 Overview of Our Method
	3 Boosting
		3.1 Problem Setting
		3.2 Our Boosting Algorithm
		3.3 Analyses
	4 ABDD Construction
	5 Circuit and SDD Construction
	6 Experiments
		6.1 Experimental Setup
		6.2 Sample-Based Robustness (SR) Validation
		6.3 Analysis
	7 Conclusion
	References
Interpretable Data Partitioning Through Tree-Based Clustering Methods
	1 Introduction
	2 Related Works
	3 Partitioning Tree Methods
	4 Experiments
	5 User Study
	6 Conclusion
	References
Jaccard-Constrained Dense Subgraph Discovery
	1 Introduction
	2 Preliminary Notation and Problem Definition
	3 Computational Complexity
	4 Algorithms
	5 Related Work
	6 Experimental Evaluation
	7 Concluding Remarks
	References
RIMBO - An Ontology for Model Revision Databases
	1 Introduction
	2 Background and Related Work
	3 Results
		3.1 Description of RIMBO
		3.2 Demonstration
	4 Discussion and Conclusion
	5 Code and Availability
	References
Unsupervised Graph Neural Networks for Source Code Similarity Detection
	1 Introduction
	2 Related Work
		2.1 Tree-Based Code Clone Detection
		2.2 Graph Neural Network
	3 Methodology
		3.1 Parser
		3.2 Graph Neural Network
		3.3 Graph Auto-encoder
		3.4 Fragment Representation
	4 Experiments
		4.1 Clone Detection
		4.2 Training and Inference
		4.3 Precision
		4.4 Analysis of Results on DNSJava
	5 Limitations and Future Research
	6 Conclusions
	References
Time Series and Forecasting
A Universal Approach for Post-correcting Time Series Forecasts: Reducing Long-Term Errors in Multistep Scenarios
	1 Introduction
		1.1 Research Questions
		1.2 Contribution
	2 Proposed Method: Time Series Post-correction (TSPC)
		2.1 Overview
		2.2 Definitions
		2.3 Nearest Shape Search
		2.4 Z-Normalization of Sliding Windows
		2.5 Meta-learning for Predicting the Success of Post-correction
		2.6 Random Forest Meta-learner
	3 Experimental Evaluation
		3.1 Datasets
		3.2 Experimental Configuration
		3.3 Evaluation Metrics
	4 Result
		4.1 Sensitivity Analysis
		4.2 Analysis of the Results
		4.3 Post-correction Meta-learner
		4.4 Persistent Pattern Among Different Datasets
		4.5 Improving Original Forecast
		4.6 Importance of the Distribution Parameter
		4.7 Effect of Tsfresh Features on the Meta-learner
	5 Related Work
	6 Conclusion
	7 Future Work
	References
Explainable Deep Learning-Based Solar Flare Prediction with Post Hoc Attention for Operational Forecasting
	1 Introduction
	2 Related Work
	3 Data and Model
	4 Interpretation Methods
	5 Experimental Evaluation
		5.1 Experimental Settings
		5.2 Model Evaluation
		5.3 Model Interpretation
	6 Conclusion and Future Work
	References
Pseudo Session-Based Recommendation with Hierarchical Embedding and Session Attributes
	1 Introduction
	2 Related Work
	3 Preliminaries
		3.1 Heterogeneous Hypergraph and Global Graph
	4 Proposed Method
		4.1 Two-Step Embedding with Category Hierarchy
		4.2 Embedding of Global Graph
		4.3 Embedding Feature Nodes
		4.4 Feature Extraction Considering Session Attributes
		4.5 Predicting and Learning About the Next Item
	5 Experiments
		5.1 Preprocessing
		5.2 Evaluation Criteria
		5.3 Comparative Model
		5.4 Parameter Setting
	6 Results and Discussion
		6.1 Performance Comparison
		6.2 Impact of Each Model Extension
	7 Conclusion
	References
Healthcare and Biological Data Analysis
Chance and the Predictive Limit in Basketball (Both College and Professional)
	1 Introduction
	2 Identifying the Impact of Chance by Monte Carlo Simulations
	3 Limitations of the MC Simulation for NCAA Basketball
	4 Related Work
	5 Deriving Limits for Specific Seasons
		5.1 Clustering Team Profiles and Deriving Match-Up Settings
		5.2 Estimating Chance
	6 Simulating Seasons
	7 Finding a Good Clustering
	8 NBA Results
	9 Summary and Conclusions
	A  Clustered schedules for different seasons, unconstrained EM
	References
Exploring Label Correlations for Quantification of ICD Codes
	1 Introduction
	2 ICD Coding of Clinical Text
		2.1 Clinical Text Dataset Splits
		2.2 Neural Network Architecture
	3 Quantification
	4 Experimental Evaluation
		4.1 Experimental Setup
		4.2 Evaluation Metrics
		4.3 Results and Discussion
	5 Related Work
	6 Conclusions and Future Work
	References
LGEM+: A First-Order Logic Framework for Automated Improvement of Metabolic Network Models Through Abduction
	1 Introduction
	2 Methods
		2.1 The First-Order Logic Framework
		2.2 Assessing Growth and Production of Compounds
		2.3 Abduction of Hypotheses
		2.4 Constraining Flux Balance Analysis Simulations Using Proofs
		2.5 Sources of Knowledge
	3 Results
	4 Discussion and Conclusion
	References
Predicting Age from Human Lung Tissue Through Multi-modal Data Integration
	1 Introduction
	2 Results
		2.1 Data Retrieval and Code Availability
		2.2 Predicting Age from Gene Expression
		2.3 Predicting Age from Methylation Arrays
		2.4 Predicting Age from Histological Images
		2.5 Ensemble Prediction Model
		2.6 Aging and Telomere Length
	3 Discussion
	4 Conclusion
	References
Anomaly, Outlier and Novelty Detection
Error Analysis on Industry Data: Using Weak Segment Detection for Local Model Agnostic Prediction Intervals
	1 Introduction
	2 Error Analysis
	3 Model Agnostic Prediction Intervals (MAPIE)
	4 Related Work
	5 Methodology
	6 Smart Pairing Use-Case
		6.1 Use-Case Description
		6.2 Experimental Results
	7 Conclusion
	8 Future Work and Limitations
	References
HEART: Heterogeneous Log Anomaly Detection Using Robust Transformers
	1 Introduction
	2 Related Work
		2.1 Log Anomaly Detection with Templates and Transformers
		2.2 Transfer Learning for Log Anomaly Detection
	3 HEART Framework, LogAnBERT and LogBERTa
		3.1 Special-Purpose Tokenizers
		3.2 LogAnBERT
		3.3 LogBERTa
		3.4 HEART Framework
	4 Experimental Setup
		4.1 Datasets
		4.2 Scenarios and Motivation
		4.3 Hardware Setup and Evaluation Metrics
	5 Results and Discussion
		5.1 Intra-system Results
		5.2 Cross-System Results
		5.3 Discussion
	6 Conclusion
	References
Multi-kernel Times Series Outlier Detection
	1 Introduction
	2 Related Work
	3 Background
		3.1 Definitions
		3.2 Support Vector Data Description (SVDD)
		3.3 Global Alignment Kernels (GAK)
	4 Multi-kernel Time Series Outlier Detection
		4.1 Fast Fourier Transform Kernels
		4.2 MK-TSOD Algorithm
		4.3 Complexity Analysis
	5 Experiments
		5.1 Setup
		5.2 Results
		5.3 Ablation Analysis
	6 Conclusions
	References
Toward Streamlining the Evaluation of Novelty Detection in Data Streams
	1 Introduction
	2 Background
		2.1 Concept Definitions
		2.2 Novelty Detection Task
		2.3 Analysis of Metrics
	3 Proposed Framework
		3.1 Novel Metric for the Binary Scenario
		3.2 Facing the Undesirable Properties of AIC
		3.3 Inclusion of the Temporal Aspect of DSs
		3.4 Inclusion of Intrinsic Data Characteristics of DSs
	4 Experiments
		4.1 Methodology
		4.2 Baseline
		4.3 Time Between the Appearance of Novel Classes
		4.4 Ratio of Offline Samples
		4.5 Number of Known Classes
	5 Conclusion
	References
Author Index