The Application of Artificial Intelligence: Step-by-Step Guide from Beginner to Expert 9783030600310, 9783030600327

1,214 267 37KB

English Pages [448] Year 2021

Report DMCA / Copyright

DOWNLOAD FILE

The Application of Artificial Intelligence: Step-by-Step Guide from Beginner to Expert
 9783030600310, 9783030600327

Table of contents :
Preface
Support Material and Software
Acknowledgments
Contents
Abbreviations
List of Figures
List of Tables
Part I: Introduction
Chapter 1: An Introduction to Machine Learning and Artificial Intelligence (AI)
1.1 Introduction
1.2 Understanding Machine Learning
1.2.1 Accuracy and Generalization Error
1.3 Supervised Learning
1.3.1 Supervised Learning Applications
1.4 Unsupervised Learning
1.4.1 Unsupervised Learning Applications
1.5 Reinforcement Learning
1.5.1 Reinforcement Learning Applications
Part II: An In-Depth Overview of Machine Learning
Chapter 2: Machine Learning Algorithms
2.1 Introduction
2.2 Supervised Learning Algorithms
2.2.1 Support Vector Machines (SVMs)
2.2.2 Feedforward Neural Networks: Deep Learning
2.2.2.1 The Activation Function
2.2.2.2 Neural Network Layer and Connection Types
2.2.2.3 The Learning Process
2.2.2.4 How to Design a Neural Network
2.2.3 Feedforward Convolutional Neural Networks: Deep Learning
2.2.3.1 Input Layer
2.2.3.2 Convolution Layer
Filter Size (W x H)
Stride (S)
Padding (P)
Calculating the Output Size of a Convolutional layer
2.2.3.3 Pooling Layer
2.2.4 Recurrent Neural Networks
2.2.4.1 LSTM Cells
2.2.5 Random Forest (Decision Tree)
2.3 Unsupervised Learning Algorithms
2.3.1 k-Means Clustering
2.3.2 MeanShift Clustering
2.3.3 DBScan Clustering
2.3.4 Hierarchical Clustering
2.4 Reinforcement Learning Algorithms
2.4.1 Action Selection Policy: How Does the Agent Select an Action?
2.4.2 Reward Function: What Is the Cumulative Future Reward?
2.4.3 State Model: How Does the Environment Behave?
2.4.4 Example 1: Simple Business Process Automation
2.4.4.1 Actions
2.4.4.2 Design of a Reward Strategy
2.4.4.3 The Learning Process
2.4.5 Example 2: Cart-Pole (Inverted Pendulum)
2.4.5.1 Actions
2.4.5.2 Environment
2.4.5.3 Design of a Reward Strategy
2.4.5.4 The Learning Process
2.5 Hybrid Models: From Autoencoders to Deep Generative Models
2.5.1 The Autoencoder (AE)
2.5.2 The Variational Autoencoder (VAE) and Generative ML Models
2.5.3 Generative Adversarial Network (GAN)
Chapter 3: Performance Evaluation of Machine Learning Models
3.1 Introduction
3.2 Performance Measures of Supervised Learning
3.2.1 RMSE
3.2.2 The Confusion Matrix
3.2.3 Accuracy
3.2.4 Cohen´s Kappa
3.2.5 Single Class Performance Measures
3.2.5.1 Precision
3.2.5.2 Recall and FNR
3.2.5.3 TNR and FPR
3.2.5.4 F1 Score
3.2.5.5 Weighted Global Performance Measures
3.2.5.6 The ROC Curve and the AUC Performance Measure
3.3 Performance Measures of Unsupervised Learning (Clustering)
3.3.1 Internal Criterion Based Performance Measures
3.3.1.1 The Silhouette Coefficient
3.3.1.2 The Calinski-Harabasz Index
3.3.1.3 The Xu-Index
3.3.1.4 Determining the Optimal Number of Clusters
3.3.2 External Criterion Based Performance Measures
3.3.2.1 Purity Measure
3.3.2.2 Contingency Table
3.3.2.3 Rand Index
3.3.2.4 Precision, Recall and F1-measure
3.3.2.5 External Criterion Based Performance Measures: Practical Example
Chapter 4: Machine Learning Data
4.1 Introduction
4.2 Data Strategy
4.3 Machine Learning Data Strategy and Tasks
4.3.1 Machine Learning (ML) Workflow
4.3.2 Data Pre-Processing
4.3.2.1 Data Cleaning
4.3.2.2 Data Transformation
4.3.2.3 Data Discretization
4.3.2.4 Data Sampling
4.3.2.5 Example: The Introduction of a New Bank Service: Sample Size
4.3.2.6 Data Resampling to Remove Class Imbalance in the Data
4.3.2.7 Feature Selection
4.3.2.8 Data Normalization: Scaling and Shifting
4.3.2.9 Training/Test Data Selection
4.3.3 Data Acquisition and Storage
4.3.3.1 Data Collection
4.3.3.2 Data Transfer
Part III: Automatic Speech Recognition
Chapter 5: Automatic Speech Recognition
5.1 Introduction
5.2 Signal Processing: The Acoustic Signal
5.2.1 The Pitch
5.2.2 The Spectrogram
5.3 Feature Extraction and Transformation
5.3.1 Feature Transformation
5.3.1.1 Linear Discriminant Analysis (LDA) Transform
5.3.1.2 Maximum Likelihood Linear Transform (MLLT)
5.3.1.3 Speaker Adaptive Training (SAT)
5.4 Acoustic Modeling
5.5 The Language Model (N-gram)
5.5.1 The Back-off (Lower Order) N-Gram Estimate
5.5.2 Unknown Words (UNK or OOV)
5.6 Pronunciation
5.7 Decoding
5.8 The Accuracy of the Trained ASR Model (WER, SER)
5.9 Practical Training of the ASR Model
Part IV: Biometrics Recognition
Chapter 6: Face Recognition
6.1 Introduction
6.2 Training a Machine Learning Model for Feature Extraction
6.3 Face Recognition with the Trained Model
6.3.1 Face Detection and Extraction
6.3.2 Face Normalization
6.3.3 Feature Extraction and Face Recognition
Chapter 7: Speaker Recognition
7.1 Introduction
7.2 Feature/Voiceprint Extraction
7.3 Feature Matching and Modeling
7.3.1 Template Modeling of Feature Vectors
7.3.2 Stochastic Modeling of Feature Vectors
7.3.3 Factor Analysis + ML Modeling of Feature Vectors (I-Vectors)
7.3.4 Neural Network Embeddings Modeling of Feature Vectors
Part V: Machine Learning by Example
Chapter 8: Machine Learning by Example
8.1 Introduction
8.2 How to Automatically Classify (Cluster) a Dataset
8.2.1 The Datasets
8.2.2 Clustering
8.2.3 Analysis of the Results
8.2.3.1 ChainLink Dataset
8.2.3.2 EngyTime Dataset
8.2.3.3 Lsun Dataset
8.2.3.4 Target Dataset
8.2.3.5 TwoDiamonds Dataset
8.2.4 General Conclusions
8.3 Dimensionality Reduction with Principal Component Analysis (PCA)
8.3.1 How to Apply PCA with the AI-TOOLKIT
8.3.1.1 Start a New Project
8.3.1.2 Create a New Database and Import the Data
8.3.1.3 Train the Model (Apply PCA)
8.3.2 Analysis of the Results
8.4 Recommendation AI
8.4.1 Collaborative Filtering (CF) Model
8.4.1.1 The User´s Feedback
8.4.1.2 How Does the Collaborative Filtering (CF) Model Work?
8.4.1.3 Evaluating the Accuracy of the Recommendation Model
8.4.1.4 Context-sensitive Recommendations
8.4.1.5 Incorporating Item Properties and User Properties
8.4.1.6 AI-TOOLKIT Collaborative Filtering (CF) Models
8.4.2 Content-Based (CB) Recommendation Model
8.4.2.1 Feature Extraction and Numerical Representation
8.4.3 Example: Movie Recommendation
8.4.3.1 Step 1: Create the Project File
8.4.3.2 Step 2: Create the AI-TOOLKIT Database and Import the Data
8.4.3.3 Step 3: Finalize the Project File
8.4.3.4 Step 4: Train the Recommendation Model
8.4.3.5 Step 5: Make a Recommendation
8.4.3.6 Analysis of the Results
8.4.3.7 Example: Recommendation for User with ID 46470
8.5 Anomaly Detection and Root Cause Analysis
8.5.1 Supervised Learning Based Anomaly Detection and Root Cause Analysis
8.5.2 Semi-supervised Learning Based Anomaly Detection
8.5.3 Unsupervised Learning Based Anomaly Detection
8.5.4 Special Considerations in Anomaly Detection
8.5.5 Data Collection and Feature Selection
8.5.6 Determining an Unknown Root Cause
8.5.7 Example: Intrusion Detection in a Military Environment (U.S. Air Force LAN): Anomaly Detection and Root Cause Analysis
8.5.7.1 Training a Machine Learning Model for the Detection of a LAN Intrusion and its Root Cause (AI-TOOLKIT)
8.6 Engineering Application of Supervised Learning Regression
8.6.1 Parameter Optimization
8.6.2 Analyzing the Results
8.7 Predictive Maintenance by Using Machine Learning
8.7.1 Data Collection
8.7.1.1 Failure History: Time Series Data
8.7.1.2 Maintenance History: Time Series Data
8.7.1.3 Machine Operating Conditions and Usage: Time Series Data
8.7.1.4 Machine Properties: Static Data
8.7.1.5 Operator Properties: Static Data
8.7.2 Feature Engineering
8.7.2.1 Aggregating Data in a Time Window
8.7.3 Labeling the Data
8.7.4 Splitting the Final Dataset into Training and Test Sets
8.7.5 Imbalance in the Data
8.7.6 Example: Predictive Maintenance of Hydropower Turbines
8.7.6.1 The Input Data: Feature Engineering and Labeling
8.7.6.2 The Machine Learning Model and Performance Evaluation
8.8 Image Recognition with Machine Learning
8.8.1 The Input Data
8.8.2 The Machine Learning Model
8.8.3 Training and Evaluation
8.9 Detecting Future Cardiovascular Disease (CVD) with Machine Learning
8.9.1 The Dataset
8.9.2 Modeling and Parameter Optimization
8.9.3 Analysis of the Results
8.10 Business Process Improvement with Machine Learning
8.10.1 The Dataset
8.10.2 Training the Machine Learning Model
8.10.3 Analysis of the Results
8.11 Replacing Measurements with Machine Learning
8.11.1 The Body Fat Dataset
8.11.2 Training the Body Fat Machine Learning Model
8.11.3 Analysis of the Results
Part VI: The AI-TOOLKIT Machine Learning Made Simple
Chapter 9: The AI-TOOLKIT: Machine Learning Made Simple
9.1 Introduction
9.2 AI-TOOLKIT Professional
9.2.1 Introduction
9.2.2 AI Definition Syntax
9.2.3 Model Types and Parameters
9.2.3.1 Supervised Learning: Support Vector Machine (SVM) Model
9.2.3.2 Supervised Learning: Random Forest Classification Model (RF)
9.2.3.3 Supervised Learning: Feedforward Neural Network Regression Model (FFNN1_R)
9.2.3.4 Supervised Learning Feedforward Neural Network Classification Model (FFNN1_C)
9.2.3.5 Supervised Learning Convolutional Feedforward Neural Network Classification Model Type 2 (FFNN2_C)
9.2.3.6 Unsupervised Learning K-Means Classification Model
9.2.3.7 Unsupervised Learning MeanShift Classification Model
9.2.3.8 Unsupervised Learning DBScan Classification Model
9.2.3.9 Unsupervised Learning Hierarchical Classification Model
9.2.3.10 Reinforcement Learning (RL): Deep Q-Learning
9.2.3.11 Dimensionality Reduction with Principal Component Analyzes (PCA)
9.2.3.12 Recommendation with Explicit Feedback (Collaborative Filtering: CFE)
9.2.3.13 Recommendation with Implicit Feedback (Collaborative Filtering: CFI)
9.2.4 Training, Testing and Prediction
9.2.5 AI-TOOLKIT Continuous Operation Module (MLFlow)
9.2.6 The AI-TOOLKIT Database
9.2.6.1 The Database Editor
9.2.6.2 Importing Data (Numerical or Categorical) into the Database
9.2.6.3 Importing Images into the Database
9.2.7 Automatic Face Recognition
9.2.7.1 How to Use
9.2.8 Automatic Speaker Recognition
9.2.8.1 How to Use the Supervector Speaker Recognition Model
9.2.8.2 How to Use the GMM and i-Vector Speaker Recognition Models
9.2.8.3 Output Options
9.2.9 Automatic Fingerprint Recognition
9.2.9.1 How to Use
9.2.9.2 The Image Controls
9.2.9.3 Settings
9.2.10 The Audio Editor
9.2.10.1 Settings
9.2.10.2 Transform Audio
9.2.10.3 Suppress Noise and Echo Cancellation
9.2.10.4 Change Pitch
9.2.10.5 Removing Audio Without Human Voice
9.2.10.6 Audio File Conversion
9.3 DeepAI Educational
9.3.1 Data Generator
9.3.2 Import External Data
9.3.3 Input Functions (Activation)
9.3.4 Layers and Nodes
9.3.5 Training the Machine Learning Model
9.3.6 Predictions
9.3.7 Settings
9.4 VoiceBridge
9.4.1 Yes-No Example
9.4.2 Exclusive Content: Inference Only Yes-No Example
9.4.3 LibriSpeech Example
9.4.3.1 Speaker Group Separation
9.4.4 Developing Your Own ASR Software with VoiceBridge
9.5 VoiceData
9.5.1 Text Normalization
9.5.1.1 How to Normalize Text
9.5.1.2 The `en_ex´ Example Grammar
9.5.2 Export ASR Data (Audio and Transcriptions)
9.5.3 AI Text Synthesizer/Speaker
9.6 Document Summary
9.6.1 How Does It Work?
9.6.2 Creating a Summary
9.6.2.1 Frequent Words
9.6.2.2 Fine Tuning Your Language Model
9.6.3 Annotating Your PDF
9.6.4 Training a Language Model
9.6.4.1 Settings
9.6.5 Extracting Text from PDF and HTML Files
9.6.6 Combining Text Files
9.6.7 Big Text Files Editor
Appendix: From Regular Expressions to HMM
A.1 Introduction
A.2 Regular Expressions
A.3 Finite State Automaton (FSA)
A.4 Finite State Transducer (FST)
A.5 Weighted Finite State Transducer (WFST)
A.6 The Hidden Markov Model (HMM)
References
Index

Polecaj historie