Machine Learning and Knowledge Extraction (Lecture Notes in Computer Science)

✍ Scribed by Andreas Holzinger (editor), Peter Kieseberg (editor), Federico Cabitza (editor), Andrea Campagner (editor), A Min Tjoa (editor), Edgar Weippl (editor)

• Publisher: Springer
• Year: 2023
• Tongue: English
• Leaves: 335
• Category: Library

✦ Synopsis

This volume LNCS-IFIP constitutes the refereed proceedings of the 7th IFIP TC 5, TC 12, WG 8.4, WG 8.9, WG 12.9 International Cross-Domain Conference, CD-MAKE 2023 in Benevento, Italy, during August 28 – September 1, 2023.

The 18 full papers presented together were carefully reviewed and selected from 30 submissions. The conference focuses on integrative machine learning approach, considering the importance of data science and visualization for the algorithmic pipeline with a strong emphasis on privacy, data protection, safety and security.

✦ Table of Contents

Preface
Organization
Contents
About the Editors
Controllable AI - An Alternative to Trustworthiness in Complex AI Systems?
1 Introduction and Motivation
2 Background
2.1 The Explainability Problem
2.2 Trustworthy AI
2.3 The AI Act
3 Principles of Controllable AI
4 Techniques for Controllable AI
4.1 Detecting Control Loss
4.2 Managing Control Loss
4.3 Support Measures
5 Discussion
6 Conclusion and Outlook for Future Research
References
Efficient Approximation of Asymmetric Shapley Values Using Functional Decomposition
1 Introduction
1.1 Shapley Values
1.2 Asymmetric Shapley Values
1.3 PDD-SHAP
2 A-PDD-SHAP
3 Experiments
3.1 Causal Explanations of Unfair Discrimination
3.2 Evaluation on Real-World Datasets
4 Summary and Outlook
5 Conclusion
A Proof of Theorem 1
References
Domain-Specific Evaluation of Visual Explanations for Application-Grounded Facial Expression Recognition
1 Introduction
2 Related Work
3 Materials and Methods
3.1 Evaluation Framework Overview
3.2 Step 1: Data Set and Model Selection
3.3 Step 2: Model Performance Analysis
3.4 Step 3: Visual Classification Explanations
3.5 Step 4: Domain-Specific Evaluation Based on Landmarks
4 Results and Discussion
5 Conclusion
References
Human-in-the-Loop Integration with Domain-Knowledge Graphs for Explainable Federated Deep Learning
1 Introduction and Motivation
2 Background and Related Work
2.1 Explainable AI on Graph Neural Networks
2.2 Federated Learning
2.3 Knowledge Graphs
2.4 Human-in-the-Loop
3 Methods, Solutions and Implementations
3.1 Disease Subnetwork Detection
3.2 Explainability
3.3 Knowledge Graph
3.4 Federated Ensemble Learning with GNNs
3.5 interaCtive expLainable plAtform for gRaph neUral networkS (CLARUS)
4 Lessons Learned
5 Conclusion and Future Outlook
References
The Tower of Babel in Explainable Artificial Intelligence (XAI)
1 Introduction and Motivation
2 Ethics Guidelines and XAI
3 Law and XAI
3.1 GDPR
3.2 Digital Services Act (DSA)
3.3 The (Proposed) Artificial Intelligence Act (AIA)
4 Standardization and XAI
5 The Link Between Law and Standardization
6 A Proposed Solution
7 Conclusion
References
Hyper-Stacked: Scalable and Distributed Approach to AutoML for Big Data
1 Introduction
2 Background and Related Work
2.1 Problem Definition
2.2 CASH Methods
2.3 Ensemble Learning
2.4 Meta-learning
2.5 Spark
3 Hyper-Stacked: A Scalable and Distributed Approach to AutoML for Big Data
3.1 Motivation
3.2 Hyper-Stacked's Design and Workflow
4 Experimental Design
4.1 Binary Supervised Learning Problems
4.2 Experimental Setups
4.3 Experiment Speedup, Sizeup, Scaleup
5 Analysis of Results
5.1 Speedup
5.2 Sizeup
5.3 Scaleup
6 Conclusions
References
Transformers are Short-Text Classifiers
1 Introduction
2 Related Work
2.1 Sequence-Based Models
2.2 Graph-Based Models
2.3 Short Text Models
2.4 Summary
3 Selected Models for Our Comparison
3.1 Models for Short Text Classification
3.2 Top-Performing Models for Text Classification
4 Experimental Apparatus
4.1 Datasets
4.2 Preprocessing
4.3 Procedure
4.4 Hyperparameter Optimization
4.5 Metrics
5 Results
6 Discussion
6.1 Key Results
6.2 Threats to Validity
6.3 Parameter Count of Models
6.4 Generalization
7 Conclusion and Future Work
References
Reinforcement Learning with Temporal-Logic-Based Causal Diagrams
1 Introduction
2 Motivating Example
3 Related Work
4 Preliminaries
5 Temporal-Logic-Based Causal Diagrams
6 Reinforcement Learning with Causal Diagrams
6.1 Q-Learning with Early Stopping
7 Case Studies
7.1 Case Study I: Small Office World Domain
7.2 Case Study II: Large Office World Domain
7.3 Case Study III: Crossroad Domain
8 Conclusions and Discussions
References
Using Machine Learning to Generate a Dictionary for Environmental Issues
1 Introduction
1.1 Findings
1.2 This Paper
2 Background
2.1 Word2Vec – CBOW and Skip-Gram
2.2 ChatGPT
2.3 Business Text – Form 10K and Earning Call Conferences
2.4 Role of Human in the Loop
3 Bag of Words
3.1 Concepts as Ontologies Represented as Bag of Words
3.2 Example System - LIWC
3.3 Concept of Interest – Carbon Footprint
4 Word2Vec – Single Words
4.1 Data for Word2Vec
4.2 Approach
4.3 Findings
4.4 Implications
4.5 Human in the Loop
5 Comparison of Word Lists Between Word2Vec and ChatGPT
5.1 Approach
5.2 Using ChatGPT to Facilitate List Analysis
5.3 Findings
5.4 Implications
6 Building a Dictionary
6.1 Carbon Footprint
6.2 Other Environmental Dictionaries
7 Positive, Negative or Action Dictionaries
8 Summary, Contributions and Extensions
8.1 Contributions
8.2 Extensions
References
Let Me Think! Investigating the Effect of Explanations Feeding Doubts About the AI Advice
1 Motivations and Background
2 Methods
2.1 Statistical Analysis
3 Results
3.1 RQ1: Impact on Decision Performance
3.2 RQ2: Impact on Decision Confidence
3.3 RQ3: Impact on Perceived Utility
4 Discussion
5 Conclusions
References
Enhancing Trust in Machine Learning Systems by Formal Methods
1 Introduction
2 State of the Art
3 What is an “Explanation”?
3.1 A Brief Survey of the Term Explanation in the Philosophy of Science
3.2 Defining Explanation for Systems Based on Machine Learning
4 Generating an Explanation
5 Applying the Method to a Meteorological Example
5.1 Description of the Meteorological Problem
5.2 Machine Learning Approach
5.3 Constructing the Explanation
6 Conclusions
References
Sustainability Effects of Robust and Resilient Artificial Intelligence
1 Introduction
2 Research Method
3 Robust and Resilient Artificial Intelligence
4 Sustainability Effects
4.1 Direct Sustainability Effects
4.2 Sustainability Effects in Selected Application Areas
5 Conclusion
References
The Split Matters: Flat Minima Methods for Improving the Performance of GNNs
1 Introduction
2 Related Work
2.1 Searching for Flat Minima
2.2 Graph Neural Networks
3 Flat Minima Methods
4 Experimental Apparatus
4.1 Datasets
4.2 Procedure
4.3 Hyperparameters
4.4 Metrics
5 Results
6 Discussion
6.1 Key Insights
6.2 Combining up to Three Flat Minima Methods
6.3 Influence of Dataset Splits
6.4 Transductive vs. Inductive Training
6.5 Detailed Discussion of Graph-MLP
6.6 Assumptions and Limitations
7 Conclusion
A Hyperparameters
A.1 Base Models
A.2 Flat Minima Methods
B Standard Deviations of Results
References
Probabilistic Framework Based on Deep Learning for Differentiating Ultrasound Movie View Planes
1 Introduction
2 Materials and Methods
2.1 The Deep Learning Algorithm
2.2 Fetal Abdomen Dataset
3 Results
4 Weighted Voted System
5 Discussion
6 Conclusions
References
Standing Still Is Not an Option: Alternative Baselines for Attainable Utility Preservation
1 Introduction
2 Related Work
3 Attainable Utility Preservation
4 Methods
5 Experimental Design
5.1 Environments
5.2 General Settings
6 Results
6.1 Comparison to AUP
6.2 Dropping the No-Op Action
7 Discussion
8 Conclusion
8.1 Future Work
References
Memorization of Named Entities in Fine-Tuned BERT Models
1 Introduction
2 Related Work
2.1 Language Models and Text Generation
2.2 Privacy Attacks in Machine Learning
2.3 Privacy Preserving Deep Learning
3 Extracting Named Entities from BERT
3.1 Fine-Tuning
3.2 Text Generation
3.3 Evaluating Named Entity Memorization
4 Experimental Apparatus
4.1 Datasets
4.2 Procedure and Implementation
4.3 Hyperparameter Optimization
4.4 Measures
5 Results
5.1 Classification
5.2 Named Entity Memorization
6 Discussion
6.1 Key Insights
6.2 Generalization
6.3 Threats to Validity
7 Conclusion
References
Event and Entity Extraction from Generated Video Captions
1 Introduction
2 Related Work
2.1 Dense Video Captioning
2.2 Text Information Extraction and Classification
3 Semantic Metadata Extraction from Videos
3.1 Dense Video Captioning (DVC)
3.2 Event Processing
3.3 Language Processing
3.4 Entity Extraction
3.5 Property Extraction
3.6 Relation Extraction
3.7 Text Classification
4 Experimental Apparatus
4.1 Datasets
4.2 Procedure
4.3 Hyperparameter Optimization
4.4 Measures and Metrics
5 Results
5.1 Dense Video Captioning
5.2 Entity Extraction
5.3 Property Extraction
5.4 Relation Extraction
5.5 Text Classification
6 Discussion
6.1 Key Results
6.2 Threats to Validity and Future Work
7 Conclusion
References
Fine-Tuning Language Models for Scientific Writing Support
1 Introduction
2 Related Work
2.1 Pre-trained Encoder Language Models
2.2 Pre-trained Decoder Language Models
2.3 Text Classification
2.4 Sentence Transformation and Paraphrasing
2.5 Tools to Improve Writing Quality
3 Experimental Apparatus
3.1 Datasets
3.2 Preprocessing
3.3 Procedure
3.4 Hyperparameter Optimization
3.5 Measures
4 Results
5 Discussion
5.1 Key Results
5.2 Threats to Validity
5.3 Ethical Considerations
6 Conclusion
References
Author Index