Machine Learning and Principles and Practice of Knowledge Discovery in Databases. International Workshops of ECML PKDD 2022 Grenoble, France, September 19–23, 2022 Proceedings

Preface
Organization
Contents – Part I
Contents – Part II
Workshop on Data Science for Social Good (SoGood 2022)
	Workshop on Data Science for Social Good (SoGood 2022)
	Organization
	Workshop Co-chairs
	Publicity Chair
	Program Committee
Responsible AI: From Principles to Action
Gender Stereotyping Impact in Facial Expression Recognition
	1 Introduction
	2 Related Work
		2.1 Facial Expression Recognition
		2.2 Bias
	3 Methodology
		3.1 Datasets
		3.2 Demographic Relabeling
		3.3 Generation of Derivative Datasets
		3.4 Experiments
		3.5 Experimental Setup
	4 Results and Discussion
		4.1 Dataset Initial Bias
		4.2 Induced Bias Impact
	5 Conclusion
	References
A Social Media Tool for Domain-Specific Information Retrieval - A Case Study in Human Trafficking
	1 Introduction
	2 Forensic Technology for Human Trafficking
	3 Architectural Overview
	4 Profile Selection Module
		4.1 Process Flow
		4.2 Evaluation and Comparison of Different Configurations
	5 URL Extraction and Crawling Module
		5.1 Process Flow
		5.2 Evaluation and Comparison of Different Configurations
	6 Case Study
	7 Conclusions and Future Work
	References
A Unified Framework for Assessing Energy Efficiency of Machine Learning
	1 Introduction
		1.1 Contribution
		1.2 Related Work
	2 Assessing Energy Efficiency of Machine Learning
		2.1 Task Configurations and Their Environment
		2.2 Monitoring for Assessing Efficiency
		2.3 Evaluation of Efficiency
		2.4 Communicating Machine Learning Efficiency
	3 Experiments
		3.1 Experimental Setup
		3.2 Efficiency Results for Image Classification
		3.3 Efficiency Results for Training Classification Models
	4 Conclusion and Future Work
	References
Fault Detection in Wastewater Treatment Plants: Application of Autoencoders Models with Streaming Data
	1 Introduction
	2 Related Work
	3 Case Studies
		3.1 Benchmark Simulation Model No 2 - BSM2
		3.2 Faults in Dissolved Oxygen Sensor
	4 Fault Detection Using Autoencoders
		4.1 LSTM Autoencoder
		4.2 Convolutional Autoencoder
	5 Experimental Results
	6 Conclusions
	References
A Temporal Fusion Transformer for Long-Term Explainable Prediction of Emergency Department Overcrowding
	1 Introduction
	2 Literature Review
	3 Data Analysis
	4 Methods
		4.1 Study Setting and Metrics
		4.2 Models
	5 Results
	6 Conclusion
	References
Exploitation and Merge of Information Sources for Public Procurement Improvement
	1 Introduction
	2 Related Work
	3 Case Study
		3.1 Data Overview
	4 Methodology
		4.1 Problem Definition
		4.2 Data Gathering
		4.3 Technologies
	5 Results
		5.1 Data Indexing
		5.2 Search by Procurement
		5.3 Search by Contract Authority and Economic Operators\' Denomination
		5.4 Definition of the Litigation Measure with Estimation of Participation in Public Tenders
		5.5 Analysis on the Graph
	6 Conclusions and Future Work
	References
Geovisualisation Tools for Reporting and Monitoring Transthyretin-Associated Familial Amyloid Polyneuropathy Disease
	1 Introduction
		1.1 Contributions
	2 Related Work
	3 Data Preparation and Subgroup Methodology
	4 Applying Geovisualisation Techniques
	5 AmiVis as a Geovisualisation Tool
	6 Ongoing Work
	7 Conclusions and Future Work
	References
Evaluation of Group Fairness Measures in Student Performance Prediction Problems
	1 Introduction
	2 Related Work
	3 Fairness Measures
		3.1 Statistical Parity
		3.2 Equal Opportunity
		3.3 Equalized Odds
		3.4 Predictive Parity
		3.5 Predictive Equality
		3.6 Treatment Equality
		3.7 Absolute Between-ROC Area
	4 Evaluation
		4.1 Datasets
		4.2 Predictive Models
		4.3 Experimental Results
		4.4 Effect of Varying Grade Threshold on Fairness
	5 Conclusion and Outlooks
	References
Combining Image Enhancement Techniques and Deep Learning for Shallow Water Benthic Marine Litter Detection
	1 Introduction
	2 Related Work
	3 Materials and Methods
		3.1 Underwater Imagery Acquisition
		3.2 Image Processing and Labeling
		3.3 Object Detection Approach
		3.4 Image Enhancement
		3.5 Combining Image Enhancement and Deep Learning
	4 Results and Discussion
		4.1 Underwater Image Enhancement
		4.2 Comparison of Objection Detection Models on Binary Class
		4.3 Combining Image Enhancement and Object Detection
	5 Conclusion and Future Work
	References
Ethical and Technological AI Risks Classification: A Human Vs Machine Approach
	1 Introduction
	2 Human Vs Machine Classification
	3 Case Study
		3.1 Risks Extraction from Literature Review
		3.2 Humans Approach
		3.3 Machine Approach
		3.4 Methods Used in Human and Machine Analysis
	4 Results and Discussion
		4.1 Classification Task by Humans
		4.2 Classification Task by Machine
		4.3 Human Vs Machine
	5 Conclusion
	References
A Reinforcement Learning Algorithm for Fair Electoral Redistricting in Parliamentary Systems
	1 Introduction
	2 Boundary & Election Datasets
	3 Redistricter Methodology
		3.1 Selection Method
		3.2 Descent-based Local Search
		3.3 Probability Updating & Smoothing
	4 Fitness Metric
		4.1 Compactness Evaluation
		4.2 Fairness Evaluation
	5 Evaluation & Results
	6 Conclusions
	Appendix 1
	Appendix 2
	References
Study on Correlation Between Vehicle Emissions and Air Quality in Porto
	1 Introduction
	2 Related Work
		2.1 Emission Inventory Over Regions to Study Road Emission
		2.2 Geo-Spatial Frameworks for Region Decomposition
	3 Data and Methods
		3.1 Porto Taxi Data
		3.2 Sensor Data
		3.3 Feature Extraction
		3.4 Vehicle Specific Power (VSP)
		3.5 VSP as an Indication of Vehicle Emission
		3.6 H3: Uber\'s Hexagonal Hierarchical Spatial Index
		3.7 Index Levels Based on the Emission and Pollution
	4 Obtained Results
		4.1 Experiment Setup
		4.2 Discussion
	5 Conclusion and Future Work
	References
Intelligently Detecting Information Online-Weaponisation Trends (IDIOT)
	1 Introduction
	2 Related Work (Online-Weaponisation)
	3 Methodology
	4 Data Collection and Annotation Description
		4.1 Dataset Choice (KDD Step 1)
		4.2 Data Cleaning and Feature Allocation. (KDD Step 1 and Step 2)
		4.3 Sentiment Analysis Open Sourced Toolboxes (KDD Step 2)
		4.4 Experimental Design (Step 2)
		4.5 Machine Learning Model Selection (KDD Step 3 and Step 4)
	5 Experimental Results, Evaluation and Discussion (Step 4)
	6 Conclusions
	7 Challenges
	8 Future Work
	References
Workshop on New Frontiers in Mining Complex Patterns (NFMCP 2022)
	New Frontiers in Mining Complex Patterns (NFMCP 2022)
	Organization
	Program Chairs
	Program Committee
Multi-modal Terminology Management
	1 Introduction
	2 Related Work
	3 Methodology
	4 Results
		4.1 Preliminary Analysis
		4.2 Data Model Setup
		4.3 Terminology Extraction
		4.4 Terminology Validation
		4.5 Termbase Deployment and Update
	5 Conclusion
	References
Cluster Algorithm for Social Choice
	1 Introduction
	2 Majority Judgement
		2.1 Formal Aspects
		2.2 Single-Winner Majority Judgement
	3 Clustering Approach
		3.1 How Clusters Work
		3.2 K-Medoids
		3.3 Clustered Majority Judgement
		3.4 Algorithm
		3.5 Case Studies
		3.6 Case Study 2: Working Hours per Week
	References
Sentimental Analysis of COVID-19 Vaccine Tweets Using BERT+NBSVM
	1 Introduction
	2 Methods
		2.1 Sentimental Classification Framework
		2.2 Collection and Pre-processing of Data
		2.3 Finding Values of Sentiments Polarity
		2.4 Combining BERT and Naive Bayes-SVM for Sentimental Classification
	3 Results and Discussion
		3.1 Sentiment Polarity
		3.2 Sentimental Classification
	4 Conclusion
	References
Rules, Subgroups and Redescriptions as Features in Classification Tasks
	1 Introduction
	2 Notation and Related Work
		2.1 Notation and Definition
		2.2 Related Work
	3 The DAFNE Framework
		3.1 Parameters Used in DAFNE Components
		3.2 Use Case Scenario
	4 Data Description
	5 Experiments and Results
	6 Conclusion and Future Work
	References
Bitpaths: Compressing Datasets Without Decreasing Predictive Performance
	1 Introduction
	2 Preliminaries
	3 Bitpaths
		3.1 Feature Construction
		3.2 Inference
		3.3 Compression
	4 Experimental Evaluation
		4.1 Experimental Evaluation Bitpaths (Q1)
		4.2 Compression Versus Accuracy (Q2)
	5 Conclusion
	References
3D Detection of ALMA Sources Through Deep Learning
	1 Introduction
	2 Methodologies
		2.1 The Pipeline
		2.2 Blobs Finder
		2.3 The Deep GRU
	3 Experiments
		3.1 Source Detection
	4 Conclusions
	References
Anomaly Detection for Physical Threat Intelligence
	1 Introduction
	2 Spark-GHSOM
	3 Spark-GHSOM for Anomaly Detection
	4 Experiments
	5 Conclusions
	References
Workshop on eXplainable Knowledge Discovery in Data Mining (XKDD 2022)
	International Workshop on eXplainable Knowledge Discovery in Data Mining (XKDD 2022)
	Organization
	XKDD 2022 Program Chairs
	XKDD 2022 Program Committee
	XKDD 2022 Keynotes: Extended Abstracts
From Attribution Maps to Concept-Level Explainable AI
The Relationship Between Explainability and Privacy in AI
Is Attention Interpretation? A Quantitative Assessment on Sets
	1 Introduction
	2 Importance Attribution as a Binary Classification Task
		2.1 Multiple-Instance Learning and Its Extensions
		2.2 Quantifying Key Instance Attribution
	3 Methods
		3.1 Attention-Based Deep MIL
		3.2 Synthetic Datasets
	4 Results
		4.1 Models with High Accuracy Can Have Poorly Behaved Attention
		4.2 Repetitions of the Same Model Have Little Correlation Between Performance and Interpretability
		4.3 Ensembling Improves Explanation Robustness
	5 Discussion
	6 Conclusion
	A  Hyperparameter Searches
		A.1  Gaussian Data
		A.2  Image Data
		A.3  CyTOF Data
	B  IAUC Distributions for Top Models
	C  Correlations Between IAUC and Accuracy
	D  Ensembling
	References
From Disentangled Representation to Concept Ranking: Interpreting Deep Representations in Image Classification Tasks
	1 Introduction and Overview
	2 Related Work
	3 Extracting Concepts for Action Classification
	4 Experimental Evaluation
	5 Ongoing Work
	References
RangeGrad: Explaining Neural Networks by Measuring Uncertainty Through Bound Propagation
	1 Introduction
	2 Related Work
		2.1 Saliency Maps
		2.2 Interval-Bound Propagation
	3 Problem Statement
	4 RangeGrad
		4.1 Interval Bound Propagation
		4.2 Rescaling
	5 Experiments
		5.1 Method Comparison
		5.2 Scaling Factor Impact
		5.3 Performance on Deeper Networks
		5.4 Sanity Checks
	6 Conclusion
	References
An Empirical Evaluation of Predicted Outcomes as Explanations in Human-AI Decision-Making
	1 Introduction
	2 Related Work
		2.1 Explainable AI
		2.2 Reliance in Human-AI Decision-Making
	3 On the Relationship of Reliance and Human-AI Decision-Making Accuracy
	4 Study Design
		4.1 Hypotheses
		4.2 Preliminaries
		4.3 Experimental Design
		4.4 Study Participants
	5 Results
		5.1 Reliance on AI Recommendations
		5.2 Human-AI Decision-Making Accuracy
	6 Discussion and Outlook
	References
Local Multi-label Explanations for Random Forest
	1 Introduction
	2 Related Work
	3 LionForests
	4 LionForest Multi-label Explainability
		4.1 Explaining Each Predicted Label Separately
		4.2 Explaining All the Predicted Labelset
		4.3 Explaining Frequent Label Subsets
	5 Experiments
		5.1 Data Sets
		5.2 Quantitative Experiments
		5.3 Qualitative Experiments
	6 Conclusion
	References
Interpretable and Reliable Rule Classification Based on Conformal Prediction
	1 Introduction
	2 Classification
	3 Decision Rules
	4 Conformal Prediction
		4.1 Transductive and Inductive Conformal Prediction
		4.2 Mondrian Conformal Prediction
	5 Conformal Decision Rules
	6 Experiments and Results
		6.1 Data Sets
		6.2 Experimental Settings
		6.3 Algorithm Output
		6.4 Results
	7 Conclusion
	References
Measuring the Burden of (Un)fairness Using Counterfactuals
	1 Introduction
	2 Related Work
		2.1 Fairness and Bias Measurement
		2.2 Counterfactual Explainability
		2.3 Counterfactual Fairness
	3 Methodology
	4 Empirical Evaluation
		4.1 Datasets
		4.2 Results and Discussion
	5 Conclusions and Future Work
	References
Are SHAP Values Biased Towards High-Entropy Features?
	1 Introduction
	2 Simulation Study
	3 Effect of Model Tuning
	4 Real Data Analysis
	5 Related Work
	6 Discussion
	A  Appendix
	References
Simple Explanations to Summarise Subgroup Discovery Outcomes: A Case Study Concerning Patient Phenotyping
	1 Introduction
	2 Background
		2.1 XAI Methods and Healthcare
		2.2 Subgroup Discovery
		2.3 SD Algorithms and Explainability
	3 Methods
	4 Experiments
		4.1 Performance and Scalability
		4.2 Use Case: Patient Phenotype
		4.3 Human Subjective Study
	5 Discussion
		5.1 Scalability of the SD Algorithms
		5.2 Decision Trees as Explainers
		5.3 Understanding of the Subgroups by Humans
	6 Conclusions
	References
Limits of XAI Application-Grounded Evaluation: An E-Sport Prediction Example
	1 Introduction
	2 State of the Art
		2.1 Evaluation of XAI
		2.2 Application-Grounded Evaluation Methodology
	3 Methodology
	4 Results
	5 Discussion
		5.1 Future Works
	6 Conclusion
	A Translation of the Main Interface
	B Translation of the Questionnaire
	References
Improving the Quality of Rule-Based GNN Explanations
	1 Introduction
	2 Related Work
		2.1 Instance-level Methods
		2.2 Model-level Methods
		2.3 Rule-based Methods
		2.4 Limitations and Desiderata
	3 Computing Activation Rules
		3.1 Using an Exhaustive Search
		3.2 Using Beam Search
		3.3 Using Monte Carlo Tree Search
	4 Transforming Rules into Subgraphs
		4.1 Optimal Transport
		4.2 Barycenter
		4.3 Associating a Graph to a Rule
	5 Experiments
		5.1 Datasets
		5.2 Computing Rules
		5.3 Finding a Representative Graph for a Rule
	6 Conclusion
	References
Exposing Racial Dialect Bias in Abusive Language Detection: Can Explainability Play a Role?
	1 Introduction
	2 Setting the Stage
		2.1 Text Classifiers
		2.2 Post-hoc Explanation Methods
	3 Preliminary Experiments
		3.1 Dataset Description
		3.2 Methods Overview
		3.3 Local to Global Explanations Scaling
		3.4 Results
	4 Conclusion and Future Work
	References
On the Granularity of Explanations in Model Agnostic NLP Interpretability
	1 Introduction and Related Work
	2 Limits of Word-Based Black-Box Interpretability
		2.1 Distributional Shift
		2.2 Computational Complexity
	3 Sentence-Based Interpretability
	4 Fidelity Experiment
	5 Discussion
	6 Conclusion
	A Reproducibility
		A.1 The Case Against Word-Based Black-Box Interpretability
		A.2 Experiments and Analysis
	B Tabular Results for OOD Classification
	C Qualitative Evaluation
	D Complete QUACKIE Results
	References
Workshop on Uplift Modeling (UMOD 2022)
	Uplift Modeling Tutorial and Workshop (UMOD 2022)
	Organization
	UMOD 2022 Chairs
	Program Committee
Estimating the Impact of Coupon Non-usage
	1 Introduction
	2 Background and the Causal Model
	3 Modeling the User Response to Coupons
	4 Observed Estimates from Real World Data
	5 Related Work
	6 Conclusion
	References
Shrinkage Estimators for Uplift Regression
	1 Introduction
	2 Shrinkage Estimators
		2.1 James-Stein Estimator
		2.2 MSE Minimizing Estimator
	3 Uplift Estimators
		3.1 The Double Estimator
		3.2 The Uplift Estimator
		3.3 The Corrected Estimator
	4 Shrinkage Uplift Estimators
		4.1 James-Stein Estimator
		4.2 MSE Minimizing Estimator
	5 Simulations
	6 Evaluation on Real Data
		6.1 Description of the Data
		6.2 Evaluation Method
		6.3 Results
	7 Conclusions
	A Proof of Theorem 1
	References
Workshop on IoT, Edge and Mobile for Embedded Machine Learning (ITEM 2022)
	Third International Workshop on IoT, Edge, and Mobile for Embedded Machine Learning (ITEM 2022)
	Organization
	Workshop Co-organizers
	Program Chair
	Technical Program Committee
Hierarchical Design Space Exploration for Distributed CNN Inference at the Edge
	1 Introduction
	2 Related Work
	3 Evaluation Methods
		3.1 Analytical Models
		3.2 AutoDiCE Framework
	4 Multi-stage Hierarchical Design Space Exploration
	5 Experimental Evaluation
		5.1 Experimental Setup
		5.2 Experimental Results
	6 Conclusion
	References
Automated Search for Deep Neural Network Inference Partitioning on Embedded FPGA
	1 Introduction
	2 Related Work
	3 Partitioning Toolflow
		3.1 Overview
		3.2 Training and Static Analysis
		3.3 DNN Partitioning
	4 Evaluation
		4.1 Workload
		4.2 Results
	5 Conclusion and Future Work
	References
Framework to Evaluate Deep Learning Algorithms for Edge Inference and Training
	1 Introduction
	2 Related Work
	3 Overall Description of the Framework
	4 Framework Implementation and Experimental Setup
	5 Case Study: Fault Diagnosis for Rotating Machinery
		5.1 Models
		5.2 Datasets
	6 Results and Discussion
		6.1 General Results
		6.2 Inference Results
		6.3 Training Results
	7 Conclusion and Future Work
	References
Hardware Execution Time Prediction for Neural Network Layers
	1 Introduction
	2 Related Work
	3 Characterization and Model Building
		3.1 Model Creation
	4 Timing Prediction
	5 Evaluation
	6 Conclusion
	References
Enhancing Energy-Efficiency by Solving the Throughput Bottleneck of LSTM Cells for Embedded FPGAs
	1 Introduction
	2 Related Work
	3 LSTM Background and Analysis
		3.1 LSTM Model vs Layer vs Cell
		3.2 Timing
	4 Optimised LSTM Cell Design
		4.1 Parallelising the LSTM Cell
		4.2 Memory Management
	5 Evaluation
		5.1 Data Set and Training Settings
		5.2 Model Implementation
		5.3 Resource Utilisation on FPGA
		5.4 Processing Time
		5.5 Inference Power
		5.6 Comparison with the State-of-the-Art
	6 Conclusion and Outlook
	References
Accelerating RNN-Based Speech Enhancement on a Multi-core MCU with Mixed FP16-INT8 Post-training Quantization
	1 Introduction
	2 RNN Based Speech Enhancement on Multi-core MCUs
		2.1 TinyDenoiser Models
		2.2 Memory Management for RNN Deployment on the Target HW
		2.3 SW Computation Model
		2.4 Mixed FP16-INT8 Post-training Quantization
	3 Experimental Results
		3.1 Accuracy After Mixed-Precision PTQ
		3.2 RNN-Based SE Inference Performance on a Multi-core MCU
		3.3 Comparison with Other Works
	4 Conclusion
	References
LDRNet: Enabling Real-Time Document Localization on Mobile Devices
	1 Introduction
	2 Related Work
	3 Context and Methodology
		3.1 Problems and Challenges
		3.2 Task Analysis
		3.3 Network Architecture
	4 Experimental Evaluation
		4.1 Training and Inference Details
		4.2 Comparison of Accuracy
		4.3 Comparison of Inference Time
		4.4 Ablation Study
		4.5 Predictions of the Occluded Points
	5 Conclusion
	References
Author Index