Neural Networks with TensorFlow and Keras: Training, Generative Models, and Reinforcement Learning

Contents About the Author About the Technical Reviewer 1 Introduction 2 Using Tensors 2.1 Tensors and NumPy 2.1.1 Array and Scalar 2.1.2 Arrays with Different Dimensions 2.1.3 Arrays with the Same Number of Dimensions 2.2 Basic Tensor Operations 2.2.1 Creating Tensors 2.2.2 Mathematical Operations 2.2.3 Reshaping Tensors 2.3 Parallelism 2.4 Machine Learning Environment 3 How Machine Learns Using Neural Network 3.1 Components of a Neural Network 3.1.1 Neurons: The Building Blocks The Sigmoid Function The Tanh Function The ReLU (Rectified Linear Unit) Function Leaky ReLU 3.1.2 Neuron Initialization 3.2 Network Layers: Building a Hierarchy 3.3 The Optimizer and the Loss Function 3.3.1 The Loss Function Using a Function Handle Using a Class Handle 3.3.2 Loss Function for Regression 3.3.3 Mean Squared Error 3.3.4 Cosine Similarity 3.4 Probabilistic Losses 3.4.1 Binary Cross-Entropy 3.4.2 Categorical Cross-Entropy 3.4.3 Sparse Categorical Cross-Entropy 3.5 Network Optimizer Time-Based Decay Step Decay Exponential Decay 3.6 Generalization Errors 3.7 TensorBoard Key Features of TensorBoard 3.8 Using TensorBoard in Colab 4 Network Layers 4.1 Dense (Fully Connected) Layers 4.2 Normalization Layers 4.3 Dropout Layers 4.3.1 Flattening Layers 4.3.2 Pooling Layers 4.3.3 Convolutional Layers 4.3.4 CNN As an Input Layer 4.3.5 Multiple CNN Layers 4.3.6 Embedding Layers 4.3.7 Residual Layers 4.3.8 Recurrent Layers 4.3.9 Activation Function 4.3.10 Recurrent Activation 4.3.11 Other Layers 5 The Training Process 5.1 Data Loading 5.1.1 Loading Images 5.2 Data Processing 5.2.1 Splitting the Dataset: Training, Development, Test 5.2.2 Categorical Data 5.2.3 Preprocessing Images 5.2.4 Normalization and Standardization 5.2.5 Missing Data 5.2.6 Data Augmentation 5.3 Tuning Our Network 5.4 Customizations 5.5 Functional API 5.6 Custom Models 5.7 Model Selection 5.8 Model Depth and Complexity 5.9 Neural Networks Applications 5.10 Dense Network: Detection of Handwritten Digits Using MNIST Dataset 5.11 RNN Network: Modeling an AutoRegressive Integrated Moving Average (ARIMA) Process 5.12 LSTM Network: BachBot 5.12.1 Background 5.12.2 Preprocessing 5.12.3 Model Implementation and Training 5.12.4 Teacher Forcing 5.12.5 BachBot Model 6 Generative Models 6.1 Variational Autoencoders 6.1.1 Preprocessing 6.1.2 VAE Architecture 6.1.3 Morphing Images 6.1.4 Feature Disentanglement 6.2 CartoonGAN 6.2.1 GAN 6.2.2 Data Preparation 6.2.3 Preprocessing CartoonGAN 6.2.4 The Discriminator Model 6.2.5 The Generator Model 6.3 Stable Diffusion 6.3.1 Text Embedding in Stable Diffusion 6.3.2 Gaussian Noise Injection and Removal 6.3.3 The U-Net Model 7 Reinforcement Learning 7.1 Explanations of Reinforcement Learning 7.2 Gymnasium Library 7.2.1 Installing Gymnasium 7.2.2 Gymnasium 7.2.3 Explaining the Gymnasium Environment Gymnasium Action and Observation Spaces Preprocessing 7.2.4 The Agent 7.2.5 Memory Replay GPU Utilization for RL 8 Using Pretrained Networks 8.1 GPT-4 8.1.1 Fine-Tuning ChatGPT 8.1.1.1 Prepare the Dataset 8.2 VGG 8.3 YOLO 8.3.1 Converting YOLO Weights to Keras 8.4 Hugging Face 8.5 Prompt Engineering 8.5.1 Zero-Shot Learning 8.5.2 Few-Shot Learning 8.5.3 One-Shot Learning 8.5.4 Chain-of-Thought Prompting 8.5.5 Role-Playing 8.5.6 Embedding Prompts 8.5.7 Knowledge Graphs 8.6 Retrieval-Augmented LLM 8.7 Best Practices for Prompt Engineering 8.7.1 Parameters 8.8 Coding an AI Agent Using LangChain 8.8.1 Indexing Using VectorDB 8.8.2 Retrieval Mechanism in LangChain 8.9 Company Chatbot Using LangChain 8.10 Other AI Agent Software 8.11 Concluding Remarks