Unlocking Artificial Intelligence: From Theory to Applications

✍ Scribed by Christopher Mutschler, Christian Münzenmayer, Norman Uhlmann, Alexander Martin

• Publisher: Springer
• Year: 2024
• Tongue: English
• Leaves: 382
• Category: Library

✦ Synopsis

Provides a concise and comprehensive overview of key areas in machine learning
Links the latest research in AI to practical applications, illustrating its real-world impact
Showcases real-world applications and how they exploit advanced analytics methods

✦ Table of Contents

Preface
Acknowledgements
Contents
Part I Theory
Chapter 1 Automated Machine Learning
1.1 Introduction
1.2 Components of AutoML Systems
1.2.1 Search Space
1.2.2 Optimization
1.2.3 Ensembling
1.2.4 Feature Selection and Engineering
1.2.5 Meta-Learning
1.2.6 A Brief Note on AutoML in the Wild
1.3 Selected Topics in AutoML
1.3.1 AutoML for Time Series Data
1.3.2 Unsupervised AutoML
1.3.3 AutoML Beyond a Single Objective
1.3.4 Human-In-The-Loop AutoML
1.4 Neural Architecture Search
1.4.1 A Brief Overview of the Current State of NAS
1.4.2 Hardware-aware NAS
1.5 Conclusion and Outlook
References
Chapter 2 Sequence-based Learning
2.1 Introduction
2.2 Time Series Processing
2.2.1 Time Series Data Streams
2.2.2 Pre-Processing
2.2.3 Predictive Modelling
2.2.4 Post-Processing
2.3 Methods
2.3.1 Temporal Convolutional Networks
2.3.2 Recurrent Neural Networks
2.3.3 Transformer
2.4 Perspectives
2.4.1 Time Series Similarity
2.4.1.1 Deep Metric Learning
2.4.2 Transfer Learning & Domain Adaptation
2.4.3 Model Interpretability
2.4.3.1 Interpretability for Time Series
2.4.3.2 Trusting Interpretations
2.5 Conclusion and Outlook
Acknowledgments
References
Chapter 3 Learning from Experience
3.1 Introduction
3.2 Concepts of Reinforcement Learning
3.2.1 Markov Decision Processes (MDPs)
3.2.2 Dynamic Programming
3.2.3 Model-free Reinforcement Learning
3.2.4 General Remarks
3.3 Learning purely through Interaction
3.3.1 Exploration-Exploitation
3.3.1.1 Exploration Strategies
3.3.1.2 Exploration in Deep RL
3.4 Learning with Data or Knowledge
3.4.1 Model-based RL with continuous Actions
3.4.2 MBRL with Discrete Actions: Monte Carlo Tree Search
3.4.3 Offline Reinforcement Learning
3.4.4 Hierarchical RL
3.5 Challenges for Agent Deployment
3.5.1 Safety through Policy Constraints
3.5.2 Generalizability of Policies
3.5.3 Lack of a Reward Function
3.6 Conclusion and Outlook
References
Chapter 4 Learning with Limited Labelled Data
4.1 Introduction
4.2 Semi-Supervised Learning
4.2.1 Classical Semi-Supervised Learning
4.2.2 Deep Semi-Supervised Learning
4.2.2.1 Self-training
4.2.2.2 Unsupervised Regularization
4.2.3 Self-Training and Consistency Regularization
4.3 Active Learning
4.3.1 Deep Active Learning (DAL)
4.3.2 Uncertainty Sampling
4.3.3 Diversity Sampling
4.3.4 Balanced Criteria
4.4 Active Semi-Supervised Learning
4.4.1 How can SSL and ALWork Together?
4.4.2 Are SSL and AL Always Mutually Beneficial?
4.5 Conclusion and Outlook
References
Chapter 5 The Role of Uncertainty Quantification for Trustworthy AI
5.1 Introduction
5.2 Towards Trustworthy AI
5.2.1 The EU AI Act
5.2.2 From Uncertainty to Trustworthy AI
5.3 Uncertainty Quantification
5.3.1 Sources of Uncertainty
5.3.1.1 Aleatoric Uncertainty
5.3.1.2 Epistemic Uncertainty
5.3.2 Methods for Quantification of Uncertainty and Calibration
5.3.2.1 Data-based Methods
5.3.2.2 Architecture-Modifying Methods
5.3.2.3 Post-Hoc Methods
5.3.3 Evaluation Metrics for Uncertainty Estimation
5.3.3.1 Negative Log-Likelihood
x
5.3.3.2 Expected Calibration Error
5.3.3.3 Rejection-based Measures
5.4 Conclusion and Outlook
References
Chapter 6 Process-aware Learning
6.1 Introduction
6.2 Overview of Process Mining
6.2.1 Process Mining Basic Concept
6.2.2 Process Mining Types
6.2.2.1 Process Discovery
6.2.2.2 Conformance Checking
6.2.2.3 Model Enhancement
6.2.3 Event Log
6.2.4 Four Quality Criteria
6.2.5 Types of Processes
6.2.5.1 Lasagna Processes
6.2.5.2 Spaghetti Processes
6.3 Process-Awareness from Theory to Practice
6.3.1 Predictive Analysis in Process Mining
6.3.2 Predictive Process Mining with Bayesian Statistics
6.3.2.1 Preliminaries for Bayesian Modeling
6.3.2.2 Quality Criteria for Bayesian Modeling
6.3.2.3 Context-Aware Structure Learning for Probabilistic Process Prediction
6.3.3 Process AI
6.4 Conclusion and Outlook
References
Chapter 7 Combinatorial Optimization
7.1 Introduction
7.2 Solving Methods
7.2.1 Heuristics
7.2.2 Exact Methods
7.3 Modeling Techniques
7.3.1 Graph Theory
7.3.1.1 Clique Problems
7.3.1.2 Flow Models
7.3.2 Mixed Integer Programs and Connections to Machine Learning
7.3.2.1 Modeling Logic
7.3.2.2 Binary Decision Trees
7.3.3 Pooling
7.4 Conclusion and Outlook
References
Chapter 8 Acquisition of Semantics for Machine-Learning and Deep-Learning based Applications
8.1 Introduction
8.2 Approaches to Acquire Semantics
8.2.1 Manual Annotation and Labeling
8.2.2 Data Augmentation Techniques
8.2.3 Simulation and Generation
8.2.3.1 Physical Modeling
8.2.3.2 Generative Adversarial Networks
8.2.4 High-End Reference Sensors
8.2.5 Active Learning
8.2.6 Knowledge Modeling Using Semantic Networks
8.2.7 Discussion
8.3 Conclusion and Outlook
References
Part II Applications
Chapter 9 Assured Resilience in Autonomous Systems – Machine Learning Methods for Reliable Perception
9.1 Introduction
9.1.1 The Perception Challenge
9.2 Approaches to reliable perception
9.2.1 Choice of Dataset
9.2.2 Unexpected Behavior of ML Methods
9.2.3 Reliable Object Detection for Autonomous Driving
9.2.4 Uncertainty Quantification for Image Classification
9.2.5 Ensemble Distribution Distillation for 2D Object Detection
9.2.6 Robust Object Detection in Simulated Driving Environments
9.2.6.1 Scenarios Setup
9.2.6.2 Methods and Metrics
9.2.6.3 Results
9.2.7 Out-of-Distribution Detection
9.3 Conclusion and Outlook
References
Chapter 10 Data-driven Wireless Positioning
10.1 Introduction
10.2 AI-Assisted Localization
10.3 Direct Positioning
10.3.1 Model
10.3.2 Experimental Setup
10.3.2.1 Measurement Campaign
10.3.2.2 Environments
10.3.3 Evaluation
10.3.4 Hybrid Localization
10.3.5 Zone Identification
10.3.6 Experimental Setup
10.3.7 Environments
10.3.8 Evaluation
10.4 Conclusion and Outlook
Acknowledgements
References
Chapter 11 Comprehensible AI for Multimodal State Detection
11.1 Introduction
11.1.1 Cognitive Load Estimation
11.1.2 Challenges in Affective Computing
11.2 Data Collection
11.2.1 Annotation
11.2.2 Data Preprocessing
11.3 Modeling
11.3.1 In-Domain Evaluation
11.3.2 Cross-Domain Evaluation
11.3.3 Interpretability
11.3.4 Improving ECG Representation Learning
11.3.5 Deployment and Application
11.4 Conclusion and Outlook
References
Chapter 12 Robust and Adaptive AI for Digital Pathology
12.1 Introduction
12.2 Applications: Tumor Detection and Tumor-Stroma Assessment
12.2.1 Generation of Labeled Data Sets
12.2.2 Data Sets for Tumor Detection
12.2.2.1 Primary Data Set
12.2.2.2 Multi-Scanner Dataset
12.2.2.3 Multi-Center Dataset
12.2.2.4 Out-of-Distribution Data Set
12.2.2.5 Urothelial Data Sets
12.2.3 Data Set for Tumor-Stroma Assessment
12.3 Prototypical Few-Shot Classification
12.3.1 Robustness through Data Augmentation
12.3.1.1 Evaluation on the Multi-Scanner Data Set
12.3.1.2 Evaluation on the Multi-Center Data Set
12.3.2 Out-of-Distribution Detection
12.3.3 Adaptation to Urothelial Tumor Detection
12.3.4 Interactive AI Authoring with MIKAIA®
12.4 Prototypical Few-Shot Segmentation
12.4.1 Tumor-Stroma Assessment
12.5 Conclusion and Outlook
Acknowledgements
References
Chapter 13 Safe and Reliable AI for Autonomous Systems
13.1 Introduction
13.1.1 Reinforcement Learning
13.1.2 Reinforcement Learning for Autonomous Driving
13.2 Generating Environments with Driver Dojo
13.2.1 Method
13.3 Training safe Policies with SafeDQN
13.3.1 Method
13.3.2 Evaluation
13.4 Extracting tree policies with SafeVIPER
13.4.1 Training the Policy
13.4.2 Verification of Decision Trees
13.4.3 Evaluation
13.5 Conclusion and Outlook
References
Chapter 14 AI for Stability Optimization in Low Voltage Direct Current Microgrids
14.1 Introduction
14.2 Low Voltage DC Microgrids
14.2.1 Control of Low Voltage DC Microgrids
14.2.2 Stability of Low Voltage DC Microgrids
14.3 AI-based Stability Optimization for Low Voltage DC Microgrids
14.3.1 Overview
14.3.2 Digital Network Twin and Generation of Labels to Describe the Stability State
14.3.3 LVDC Microgrid Surrogate Model Applying Random Forests
14.3.4 Stability Optimization Applying Decision Trees
14.4 Implementation and Assessment
14.4.1 Measurement of Grid Stability
14.4.2 Experimental Validation
14.5 Conclusion and Outlook
References
Chapter 15 Self-Optimization in Adaptive Logistics Networks
15.1 Introduction
15.2 A Brief Overview of Relevant Literature on Predicting the All-Time Buy Quantity
15.3 Predicting the All-Time Buy
15.4 A Probabilistic Hierarchical Growth Curve model
15.5 Determining the Optimal Order Policy
15.5.1 Modeling Non-Linear Costs
15.5.2 Robust Optimization
15.6 Pooling
15.7 Conclusion and Outlook
References
Chapter 16 Optimization of Underground Train Systems
16.1 Optimization of DC Railway Power Systems
16.1.1 Introduction
16.1.2 Optimal Power Flow and mathematical MIQCQP model
16.1.2.1 Snapshot Model
16.1.2.2 Time Span Model
16.1.3 Case Studies
16.1.3.1 Optimization of time stamps in a small network
16.1.3.2 Optimization of a realistic entire line
16.2 Energy-Efficient Timetabling applied to a German Underground System
16.2.1 Industrial Challenge and Motivation
16.2.2 Mathematical Research
16.2.3 Implementation
16.3 Conclusion and Outlook
Acknowledgements
References
Chapter 17 AI-assisted Condition Monitoring and Failure Analysis for IndustrialWireless Systems
17.1 Introduction
17.2 Verifying Data Source Accuracy in Protocol Analysis
17.2.1 System Concept
17.2.2 Autoencoder Architecture for Anomaly Detection
17.2.3 Dataset and Performance Evaluation
17.3 Automated and User-friendly Spectral Analysis
17.3.1 ML-based Spectrum Analysis
17.3.2 Generation of Training and Validation Data
17.3.3 Model Validation Using Artificial and Measurement Data
17.3.4 System Architecture
17.4 Cross-layer Analysis
17.4.1 Variable Adaptive Dynamic Time Warping: A Novel Approach
17.4.2 Experimental Results and Discussion
17.4.3 Implications for Research and Beyond
17.5 Conclusion and Outlook
References
Chapter 18 XXL-CT Dataset Segmentation
18.1 Introduction
18.2 XXL-CT Dataset Acquisition
18.2.0.1 Me163 Airplane
18.2.0.2 Honda Accord Vehicle
18.3 Annotation Pipelines
18.3.1 3D Instance Labelling Pipeline
18.3.2 3D Semantic Labelling Pipeline
18.4 Training Infrastructure and Segmentation Results
18.4.1 Instance Segmentation
18.4.2 Semantic Segmentation
18.5 Conclusion and Outlook
Acknowledgments
References
Chapter 19 Energy-Efficient AI on the Edge
19.1 AI on the Edge
19.2 Energy-Efficient Classical Machine Learning
19.2.1 Classification of Time Series Data
19.2.2 Multi-Objective Optimization
19.2.3 Energy Prediction for Classical Machine Learning
19.2.4 EA-AutoML Tool
19.2.5 Application Example
19.3 Energy-Efficient Deep Learning
19.3.1 Deep Compression
19.3.1.1 Pruning
19.3.1.2 Quantization
19.3.2 Efficient Design Space Exploration
19.3.3 Benchmarking Edge AI
19.4 Conclusion and Outlook
References