Deep Learning Approaches for Security Threats in Iot Environments

ISBN-13: 9781119884149 / Angielski / Twarda / 2022 / 476 str.

Hossam Reda Mohamed Hawash

Spis treści

About the Authors xv1 Introducing Deep Learning for IoT Security 11.1 Introduction 11.2 Internet of Things (IoT) Architecture 11.2.1 Physical Layer 31.2.2 Network Layer 41.2.3 Application Layer 51.3 Internet of Things' Vulnerabilities and Attacks 61.3.1 Passive Attacks 61.3.2 Active Attacks 71.4 Artificial Intelligence 111.5 Deep Learning 141.6 Taxonomy of Deep Learning Models 151.6.1 Supervision Criterion 151.6.1.1 Supervised Deep Learning 151.6.1.2 Unsupervised Deep Learning 171.6.1.3 Semi-Supervised Deep Learning 181.6.1.4 Deep Reinforcement Learning 191.6.2 Incrementality Criterion 191.6.2.1 Batch Learning 201.6.2.2 Online Learning 211.6.3 Generalization Criterion 211.6.3.1 Model-Based Learning 221.6.3.2 Instance-Based Learning 221.6.4 Centralization Criterion 221.7 Supplementary Materials 25References 252 Deep Neural Networks 272.1 Introduction 272.2 From Biological Neurons to Artificial Neurons 282.2.1 Biological Neurons 282.2.2 Artificial Neurons 302.3 Artificial Neural Network 312.3.1 Input Layer 342.3.2 Hidden Layer 342.3.3 Output Layer 342.4 Activation Functions 352.4.1 Types of Activation 352.4.1.1 Binary Step Function 352.4.1.2 Linear Activation Function 362.4.1.3 Nonlinear Activation Functions 362.5 The Learning Process of ANN 402.5.1 Forward Propagation 412.5.2 Backpropagation (Gradient Descent) 422.6 Loss Functions 492.6.1 Regression Loss Functions 492.6.1.1 Mean Absolute Error (MAE) Loss 502.6.1.2 Mean Squared Error (MSE) Loss 502.6.1.3 Huber Loss 502.6.1.4 Mean Bias Error (MBE) Loss 512.6.1.5 Mean Squared Logarithmic Error (MSLE) 512.6.2 Classification Loss Functions 522.6.2.1 Binary Cross Entropy (BCE) Loss 522.6.2.2 Categorical Cross Entropy (CCE) Loss 522.6.2.3 Hinge Loss 532.6.2.4 Kullback-Leibler Divergence (KL) Loss 532.7 Supplementary Materials 53References 543 Training Deep Neural Networks 553.1 Introduction 553.2 Gradient Descent Revisited 563.2.1 Gradient Descent 563.2.2 Stochastic Gradient Descent 573.2.3 Mini-batch Gradient Descent 593.3 Gradient Vanishing and Explosion 603.4 Gradient Clipping 613.5 Parameter Initialization 623.5.1 Zero Initialization 623.5.2 Random Initialization 633.5.3 Lecun Initialization 653.5.4 Xavier Initialization 653.5.5 Kaiming (He) Initialization 663.6 Faster Optimizers 673.6.1 Momentum Optimization 673.6.2 Nesterov Accelerated Gradient 693.6.3 AdaGrad 693.6.4 RMSProp 703.6.5 Adam Optimizer 703.7 Model Training Issues 713.7.1 Bias 723.7.2 Variance 723.7.3 Overfitting Issues 723.7.4 Underfitting Issues 733.7.5 Model Capacity 743.8 Supplementary Materials 74References 754 Evaluating Deep Neural Networks 774.1 Introduction 774.2 Validation Dataset 784.3 Regularization Methods 794.3.1 Early Stopping 794.3.2 L1 and L2 Regularization 804.3.3 Dropout 814.3.4 Max-Norm Regularization 824.3.5 Data Augmentation 824.4 Cross-Validation 834.4.1 Hold-Out Cross-Validation 844.4.2 k-Folds Cross-Validation 854.4.3 Stratified k-Folds' Cross-Validation 864.4.4 Repeated k-Folds' Cross-Validation 874.4.5 Leave-One-Out Cross-Validation 884.4.6 Leave-p-Out Cross-Validation 894.4.7 Time Series Cross-Validation 904.4.8 Rolling Cross-Validation 904.4.9 Block Cross-Validation 904.5 Performance Metrics 924.5.1 Regression Metrics 924.5.1.1 Mean Absolute Error (MAE) 924.5.1.2 Root Mean Squared Error (RMSE) 934.5.1.3 Coefficient of Determination (R²) 934.5.1.4 Adjusted R² 944.5.2 Classification Metrics 944.5.2.1 Confusion Matrix 944.5.2.2 Accuracy 964.5.2.3 Precision 964.5.2.4 Recall 974.5.2.5 Precision-Recall Curve 974.5.2.6 F1-Score 974.5.2.7 Beta F1 Score 984.5.2.8 False Positive Rate (FPR) 984.5.2.9 Specificity 994.5.2.10 Receiving Operating Characteristics (ROC) Curve 994.6 Supplementary Materials 99References 1005 Convolutional Neural Networks 1035.1 Introduction 1035.2 Shift from Full Connected to Convolutional 1045.3 Basic Architecture 1065.3.1 The Cross-Correlation Operation 1065.3.2 Convolution Operation 1075.3.3 Receptive Field 1085.3.4 Padding and Stride 1095.3.4.1 Padding 1095.3.4.2 Stride 1115.4 Multiple Channels 1135.4.1 Multi-Channel Inputs 1135.4.2 Multi-Channel Output 1145.4.3 Convolutional Kernel 1 × 1 1155.5 Pooling Layers 1165.5.1 Max Pooling 1175.5.2 Average Pooling 1175.6 Normalization Layers 1195.6.1 Batch Normalization 1195.6.2 Layer Normalization 1225.6.3 Instance Normalization 1245.6.4 Group Normalization 1265.6.5 Weight Normalization 1265.7 Convolutional Neural Networks (LeNet) 1275.8 Case Studies 1295.8.1 Handwritten Digit Classification (One Channel Input) 1295.8.2 Dog vs. Cat Image Classification (Multi-Channel Input) 1305.9 Supplementary Materials 130References 1306 Dive Into Convolutional Neural Networks 1336.1 Introduction 1336.2 One-Dimensional Convolutional Network 1346.2.1 One-Dimensional Convolution 1346.2.2 One-Dimensional Pooling 1356.3 Three-Dimensional Convolutional Network 1366.3.1 Three-Dimensional Convolution 1366.3.2 Three-Dimensional Pooling 1366.4 Transposed Convolution Layer 1376.5 Atrous/Dilated Convolution 1446.6 Separable Convolutions 1456.6.1 Spatially Separable Convolutions 1466.6.2 Depth-wise Separable (DS) Convolutions 1486.7 Grouped Convolution 1506.8 Shuffled Grouped Convolution 1526.9 Supplementary Materials 154References 1547 Advanced Convolutional Neural Network 1577.1 Introduction 1577.2 AlexNet 1587.3 Block-wise Convolutional Network (VGG) 1597.4 Network in Network 1607.5 Inception Networks 1627.5.1 GoogLeNet 1637.5.2 Inception Network v2 (Inception v2) 1667.5.3 Inception Network v3 (Inception v3) 1707.6 Residual Convolutional Networks 1707.7 Dense Convolutional Networks 1737.8 Temporal Convolutional Network 1767.8.1 One-Dimensional Convolutional Network 1777.8.2 Causal and Dilated Convolution 1807.8.3 Residual Blocks 1857.9 Supplementary Materials 188References 1888 Introducing Recurrent Neural Networks 1898.1 Introduction 1898.2 Recurrent Neural Networks 1908.2.1 Recurrent Neurons 1908.2.2 Memory Cell 1928.2.3 Recurrent Neural Network 1938.3 Different Categories of RNNs 1948.3.1 One-to-One RNN 1958.3.2 One-to-Many RNN 1958.3.3 Many-to-One RNN 1968.3.4 Many-to-Many RNN 1978.4 Backpropagation Through Time 1988.5 Challenges Facing Simple RNNs 2028.5.1 Vanishing Gradient 2028.5.2 Exploding Gradient 2048.5.2.1 Truncated Backpropagation Through Time (TBPTT) 2048.5.2.2 Penalty on the Recurrent Weights Whh2058.5.2.3 Clipping Gradients 2058.6 Case Study: Malware Detection 2058.7 Supplementary Material 206References 2079 Dive Into Recurrent Neural Networks 2099.1 Introduction 2099.2 Long Short-Term Memory (LSTM) 2109.2.1 LSTM Gates 2119.2.2 Candidate Memory Cells 2139.2.3 Memory Cell 2149.2.4 Hidden State 2169.3 LSTM with Peephole Connections 2179.4 Gated Recurrent Units (GRU) 2189.4.1 CRU Cell Gates 2189.4.2 Candidate State 2209.4.3 Hidden State 2219.5 ConvLSTM 2229.6 Unidirectional vs. Bidirectional Recurrent Network 2239.7 Deep Recurrent Network 2269.8 Insights 2279.9 Case Study of Malware Detection 2289.10 Supplementary Materials 229References 22910 Attention Neural Networks 23110.1 Introduction 23110.2 From Biological to Computerized Attention 23210.2.1 Biological Attention 23210.2.2 Queries, Keys, and Values 23410.3 Attention Pooling: Nadaraya-Watson Kernel Regression 23510.4 Attention-Scoring Functions 23710.4.1 Masked Softmax Operation 23910.4.2 Additive Attention (AA) 23910.4.3 Scaled Dot-Product Attention 24010.5 Multi-Head Attention (MHA) 24010.6 Self-Attention Mechanism 24210.6.1 Self-Attention (SA) Mechanism 24210.6.2 Positional Encoding 24410.7 Transformer Network 24410.8 Supplementary Materials 247References 24711 Autoencoder Networks 24911.1 Introduction 24911.2 Introducing Autoencoders 25011.2.1 Definition of Autoencoder 25011.2.2 Structural Design 25311.3 Convolutional Autoencoder 25611.4 Denoising Autoencoder 25811.5 Sparse Autoencoders 26011.6 Contractive Autoencoders 26211.7 Variational Autoencoders 26311.8 Case Study 26811.9 Supplementary Materials 269References 26912 Generative Adversarial Networks (GANs) 27112.1 Introduction 27112.2 Foundation of Generative Adversarial Network 27212.3 Deep Convolutional GAN 27912.4 Conditional GAN 28112.5 Supplementary Materials 285References 28513 Dive Into Generative Adversarial Networks 28713.1 Introduction 28713.2 Wasserstein GAN 28813.2.1 Distance Functions 28913.2.2 Distance Function in GANs 29113.2.3 Wasserstein Loss 29313.3 Least-Squares GAN (LSGAN) 29813.4 Auxiliary Classifier GAN (ACGAN) 30013.5 Supplementary Materials 301References 30114 Disentangled Representation GANs 30314.1 Introduction 30314.2 Disentangled Representations 30414.3 InfoGAN 30614.4 StackedGAN 30914.5 Supplementary Materials 316References 31615 Introducing Federated Learning for Internet of Things (IoT) 31715.1 Introduction 31715.2 Federated Learning in the Internet of Things 31915.3 Taxonomic View of Federated Learning 32215.3.1 Network Structure 32215.3.1.1 Centralized Federated Learning 32215.3.1.2 Decentralized Federated Learning 32315.3.1.3 Hierarchical Federated Learning 32415.3.2 Data Partition 32515.3.3 Horizontal Federated Learning 32615.3.4 Vertical Federated Learning 32715.3.5 Federated Transfer Learning 32815.4 Open-Source Frameworks 33015.4.1 TensorFlow Federated 33015.4.2 PySyft and PyGrid 33115.4.3 FedML 33115.4.4 LEAF 33215.4.5 PaddleFL 33215.4.6 Federated AI Technology Enabler (FATE) 33315.4.7 OpenFL 33315.4.8 IBM Federated Learning 33315.4.9 NVIDIA Federated Learning Application Runtime Environment (NVIDIA FLARE) 33415.4.10 Flower 33415.4.11 Sherpa.ai 33515.5 Supplementary Materials 335References 33516 Privacy-Preserved Federated Learning 33716.1 Introduction 33716.2 Statistical Challenges in Federated Learning 33816.2.1 Nonindependent and Identically Distributed (Non-IID) Data 33816.2.1.1 Class Imbalance 33816.2.1.2 Distribution Imbalance 34116.2.1.3 Size Imbalance 34616.2.2 Model Heterogeneity 34616.2.2.1 Extracting the Essence of a Subject 34616.2.3 Block Cycles 34816.3 Security Challenge in Federated Learning 34816.3.1 Untargeted Attacks 34916.3.2 Targeted Attacks 34916.4 Privacy Challenges in Federated Learning 35016.4.1 Secure Aggregation 35116.4.1.1 Homomorphic Encryption (HE) 35116.4.1.2 Secure Multiparty Computation 35216.4.1.3 Blockchain 35216.4.2 Perturbation Method 35316.5 Supplementary Materials 355References 355Index 357

O autorze

Mohamed Abdel-Basset, PhD, is an Associate Professor in the Faculty of Computers and Informatics at Zagazig University, Egypt. He is a Senior Member of the IEEE.Nour Moustafa, PhD, is a Postgraduate Discipline Coordinator (Cyber) and Senior Lecturer in Cybersecurity and Computing at the School of Engineering and Information Technology at the University of New South Wales, UNSW Canberra, Australia.Hossam Hawash is an Assistant Lecturer in the Department of Computer Science, Faculty of Computers and Informatics at Zagazig University, Egypt.