Scaling Graph Learning for the Enterprise. Production-Ready Graph Learning and Inference Ahmed Menshawy, Sameh Mohamed, Maraim Rizk Masoud

Dostępność produktu

Produkt nie został jeszcze oceniony pod kątem ułatwień dostępu lub nie podano żadnych informacji o ułatwieniach dostępu lub są one niewystarczające. Prawdopodobnie Wydawca/Dostawca jeszcze nie umożliwił dokonania walidacji produktu lub nie przekazał odpowiednich informacji na temat jego dostępności.

Spis treści książki

• Preface
  • What Is Graph Learning for the Enterprise?
  • Why Does Graph Learning Matter for the Enterprise?
  • Who Is This Book For?
  • Overview of the Chapters
  • Conventions Used in This Book
  • Using Code Examples
  • OReilly Online Learning
  • How to Contact Us
  • Acknowledgements
• 1. Introduction to Graphs
  • The Power of Enterprise Graph Learning and Inference at Scale
  • A Birds-Eye View: Navigating the Books Chapters
  • Graphs and Graph Learning
    • What Is a Graph?
    • Graph Data Representation
    • Graph Learning
    • Scalable Graph Learning: Addressing the Requirements
    • Advantages of Scalable Graph Learning in Enterprise
  • Large-Scale Graphs in Real-World Enterprises: Use Cases
    • Travel-Time Predictions on Google Maps
    • Drug Development: Halicin
    • Fraud Detection
  • The Evolution of Graphs and Graph Learning: From Early Beginnings to Modern Applications
    • Era 1: The Foundation of Graph Theory and Algorithms (1736-1970)
    • Era 2: More Advancement in Graph Algorithms and Technologies (1970-1999)
    • Era 3: Emergence of Graph Databases and Graph Query Languages (2000-2006)
    • Era 4: Graph Analytics and Traditional Machine Learning (2007-2011)
    • Era 5: Rise of Graph Neural Networks (2012-2018)
    • Era 6: Scalability, Robustness, and Enterprise Applications (2019-Present)
    • Challenges of Enterprise-Ready Graph Learning Systems
    • Data Harmonization Challenges
    • Computationally Intensive Workloads
    • Dynamic Evolving Graphs
    • Active Monitoring and Drift Detection
    • Real-Time Inference
  • Summary
• 2. The Graph Machine Learning Pipeline
  • The Graph Data Pipeline
    • Definition of Graph Data and Graph Data Levels
      • Natural graph data
      • Graphically represented phenomena
      • Structured data
      • Unstructured data
    • Graph Data Sourcing and Understanding
      • Graph data sourcing
      • Understanding visual graph data
    • Graph Data Preparation
      • Graph data ingestion
      • Graph data storage
      • Graph data preprocessing
      • Why enrich graph data with external data?
      • Graph data harmonization
  • The Graph Training and Inference Pipelines
    • GML Training Pipeline Overview
      • Graph feature engineering
      • Graph model building cycle
    • GML Inference Pipeline Overview
      • GML model registration
      • GML model serving
      • Monitoring and feedback loop
  • Summary
• 3. Traditional Machine Learning for Graphs
  • Approaches to Graph Machine Learning
    • Traditional Graph-Based Machine Learning
    • Nontraditional Graph-Based Machine Learning
  • Representing Graphs for Traditional ML
    • Graph Representation
    • Representing Amazon Copurchasing Networks as Graphs
    • Navigating Graph Tasks in the Amazon Copurchasing Dataset
  • Graph Feature Engineering
    • Importance and Challenges
    • Types of Graph Features
      • Node-level features
      • Edge-level features
      • Graph-level features
    • Hands-on: Extracting Features for the Amazon Copurchasing Graph
      • Graph feature derivation
      • Leveraging graph-derived metrics to enrich product attributes
  • Graph Features in ML Modeling
    • Task and Techniques Overview
      • Node classification
      • Link prediction
      • Graph clustering
  • Predicting High-Rated Products with a Prediction Model
  • Feature Learning with Node Embeddings
    • Random Walk Algorithm
    • Amazon Copurchasing Dataset and Node Embeddings
      • Node embeddings use case for Amazon copurchasing dataset
      • Generating node embeddings
  • Summary
• 4. PyGraf: End-to-End Graph Learning and Serving
  • Graph Libraries Overview
    • Challenges of Open Source Graph Libraries: PyGraf Opportunities
    • PyGraf: A Solution for Streamlined Graph Learning and Serving
  • Introduction to PyGraf
    • PyGraf Key Features
    • PyGraf Purposes: Empowering Dynamic Environments
  • Architecture and Core Capabilities
    • Core Components Layer
    • Adaptation and Integration Layer
    • Best Practices Layer
  • In-Depth Exploration of Core Library Components
    • Data Component
      • Pre-integrated datasets
      • Extending with new datasets
      • Connectors in PyGraf
      • Preprocessors and Transformation in PyGraf
    • Training Component
      • Graph learning (Training)
      • Fine-tuning
      • Model selection and validation
    • Serving Component
      • Serving mechanisms: Scaling and optimizing for production
      • NVIDIA Triton Inference Server example
    • Privacy Preserving Component
      • Federated learning: Principles and implementation in PyGraf
      • Implementing federated learning in PyGraf
  • End-to-End Example Using PyGraf: Amazon Copurchasing Dataset
    • Preprocessing and Transformation
    • Model Training
    • Evaluation and Model Selection
    • Deployment and Monitoring
  • Summary
• 5. Graph Neural Networks
  • Introduction to Graph Neural Networks
    • The Significance of GNNs in Graph Learning
    • Overview of GNN Applications
    • Foundations of GNNs
      • Message passing in GNNs
      • Learning process
    • Graph Convolutional Networks
      • The convolution operation
      • The GCN layer operation
      • Simple example
    • Transition from Traditional Graph Learning to GCNs
    • How GCNs Simplify Feature Engineering
  • Hands-on E2E Example Using PyG and the PyGraf Interface
    • PyG: Karate Club Example
      • PyG installation
      • PyG graph data representation
      • PyG GCN API interface
      • Embedding the Karate Club network
    • Cora Node Classification
      • Data preparation
      • GCN model architecture
      • Embedding the Cora network
      • Model training and evaluation
      • Model testing
    • PyGraf API Interface
      • Initializing modules and dependencies
      • Data loading and preprocessing
      • Model initialization
      • Setting up the training components
      • Training logic
      • Model evaluation
      • Visualization of embeddings
  • Limitations of GCN-Based Architecture
  • Summary
• 6. Advanced Techniques in Graph Learning
  • Different Types of Graphs
    • Homogeneous Graphs
    • Heterogeneous Graphs
    • Temporal Graphs
  • Graph Embedding Models
    • How Do Knowledge Graph Embeddings Work?
    • Training Knowledge Graph Embedding Models
    • Embedding Interaction Methods
      • Distance-based embedding interactions
      • Factorization-based embedding interactions
    • Training Objectives
    • Strengths of Graph Embedding Models
    • Example: Learning on Freebase Dataset
      • Loading and exploring data
      • Initializing the model and training configuration
      • Defining and executing the training loop
  • Attention on Graphs
    • Feature Vectors
    • Attention Mechanism
    • Multihead Attention
    • Example: Citation Network
      • Feature combination
      • Score calculation
      • Normalization (softmax)
      • Weighted sum
      • Update paper i information
  • CiteSeer Use Case Example
    • Data Preprocessing and Visualization
    • Model Training Using PyG
    • Model Testing
    • Embeddings Visualization
  • Summary
• 7. Scalable Graph Neural Networks
  • Challenges in Scaling Graph Learning Models
    • Computational Challenges
    • Memory Constraints
    • Data Complexity and Size
  • Mini-Batching in Graph Neural Networks
    • Definition and Importance of Mini-Batching
    • Techniques for Effective Mini-Batching
      • Implementation with PyG DataLoader
      • Example: Batching pair and bipartite graphs
  • Memory-Efficient Training Techniques
    • Gradient Checkpointing
    • Subgraph Sampling Methods
    • Layer-Wise Relevance Propagation
  • Distributed Data and Compute Strategies
    • Distributed Execution Strategies
      • Conventional mini-batching
      • Factored mini-batching
      • Operator-parallel mini-batching
      • Pull-push parallelism
      • Comparing the approaches
    • Graph Partitioning Strategies
      • Node splitting
      • Edge splitting
    • Distributed Graph Learning Tools
      • PyTorch Geometric
      • PaddlePaddle: Paddle Graph Learning (PGL)
  • Distributed Training with PyG
    • Prepare and Partition the Graph Data
    • Manual Execution of Distributed Training
    • Key Components of Distributed Training with PyG
  • Advanced Architectures for Scaling Graph Neural Networks
    • Sparsity Exploitation
    • Approximation Techniques
    • Use of External Memory
  • Summary
• 8. Enterprise Applications of Graphs
  • Customer and Market Insights
    • Customer Segmentation
    • Social Network Analysis
    • Recommendation Engines
  • Operations and Supply Chain Management
    • Network Optimization in Supply Chains
    • Inventory Management
    • Risk Management and Supply Chain Resilience
      • Supplier selection and relationship management
      • Demand forecasting and distribution planning
      • Transportation and logistics optimization
    • Real-World Applications
      • Amazon: Supply chain and logistics optimization
      • Walmart: Inventory management and supplier relationships
      • UPS: Transportation and logistics
  • Security and Risk Management
    • Threat Detection and Analysis
    • Identity and Access Management
    • Fraud Detection
    • Cybersecurity Incident Response
    • Risk Management and Compliance
  • Healthcare and Life Sciences
    • Patient Journey Mapping and Personalization
    • Drug Discovery and Development
      • Accelerating drug discovery and repurposing
      • Optimizing drug safety monitoring
    • Clinical Trial Optimization
    • Genomics and Personalized Medicine
    • Popular Life Science Products
      • BenevolentAI drug repurposing
      • Tempus genomic analysis
      • Medidata Rave for clinical trials
  • Summary
• 9. Privacy-Preserving Graph Learning
  • The Importance of Privacy in Graph Learning
  • Enterprise Examples and Applications
  • Overview of Privacy-Preserving Techniques
    • Privacy Threats in Graph Learning
      • Types of privacy attacks
      • Adversarial models and background knowledge
    • Case Studies of Privacy Breaches
      • The Netflix Prize data leak
      • Social network de-anonymization
  • Privacy-Preserving Techniques for Graph Data
    • Graph Data Anonymization
    • k-Anonymity and t-Closeness in Graphs
    • Graph Modification Techniques
      • Edge perturbation
      • Node aggregation
      • Utility considerations
    • Edge Differential Privacy
    • Node Differential Privacy
    • Synthetic Graph Data Generation
      • Graph GANs
      • Variational autoencoders (VAEs)
  • Privacy-Preserving Graph Computation
    • SMPC for Graphs
      • Concept of SMPC
      • SMPC Protocols for Graph Computation
    • Homomorphic Encryption in Graph Learning
      • Concept of homomorphic encryption
      • Applying homomorphic encryption to graph learning
    • Differentially Private Learning Algorithms
      • Concept of differential privacy in learning
      • Techniques for differentially private learning
      • Applying differential privacy to graph models
      • Example: Training a GNN with differential privacy
      • Training with differential privacy considerations
  • Federated Graph Learning
    • Principles and Architecture
      • Federated learning architecture
      • Training process
      • Relevance to graph data
      • Motivation for federated graph learning
    • Federated Graph Neural Networks
      • Model architectures
      • Fintech example
    • Training and Communication Protocols
      • Federated training protocol for GNNs
      • Communication protocols
    • Privacy Enhancements in Federated Graph Learning
      • Secure aggregation protocols
      • Example of a secure aggregation protocol
      • Differential privacy in federated settings
  • Applications of Privacy-Preserving Graph Learning
    • Privacy-Preserving Techniques in Social Network Analysis
      • Anonymization of social graphs
      • Differential privacy in social networks
      • Federated social network analysis
    • Use Cases
  • Case Study: Applying FedGraphNN to a Recommender System Using Epinions Data
    • Dataset
    • Importance of Applying Federated Learning
    • Data Processing
    • Federated Training Logic
      • Model definition: GCN for link prediction
      • Trainer definition: Federated training logic
    • Running the Federated Pipeline
  • Summary
• 10. Graph Inference and Deployment Strategies
  • Deployment Strategies
    • Canary Deployments: Testing the Waters
    • BlueGreen Deployments: Smooth Transitions
    • Shadow Deployments: Flying Under the Radar
    • Edge Deployments: Taking It to the Edge
    • A/B Testing Deployments: Controlled Experiments
    • Progressive Rollouts: Scaling Up Gradually
    • Containerized Deployments
    • Hardware Considerations
      • CPU inference
      • GPU inference
      • Emerging hardware technologies
      • Balancing hardware choices
  • Inference Runtimes
    • Inference Frameworks and Libraries
    • Precomputations and Caching
      • Precomputed embeddings
      • Caching inference results
      • Trade-offs in precomputation and caching
    • Online Versus Offline Inference
      • Online inference: Real-time predictions
      • Offline inference: Batch processing
      • Balancing online and offline inference
  • Model Optimization Techniques
    • Quantization
      • How quantization works
      • Types of quantization
      • Technical considerations
    • Pruning
      • How pruning works
      • Advantages of pruning for graph models
      • Limitations and challenges of pruning
    • Knowledge Distillation
      • How knowledge distillation works
      • Advantages of knowledge distillation for graph model inference
      • Teacherstudent architectures for graph models
  • Scaling Inference for Large Graphs
    • Distributed Inference Systems
    • Graph Partitioning for Scalable Inference
      • How graph partitioning works
      • Techniques for graph partitioning
      • Challenges and trade-offs in graph partitioning
    • Incremental Inference
      • The fundamentals of incremental inference
      • Techniques in incremental inference
      • Challenges in incremental inference
      • Addressing challenges in incremental inference
  • Summary
• 11. Monitoring and Feedback Loops
  • Challenges in Monitoring Graph Models
    • Graph-Specific Metrics
    • Dynamic Graphs
    • Scale and Complexity
    • System and Deployment Issues
  • Designing a Monitoring Framework
    • Dealing with Ever-Changing Connections (Graph Topology)
    • Checking the Meaning (Evaluating Embeddings)
    • The Balancing Act (System Performance Versus Graph Tasks)
    • Data Collection Mechanisms
      • Real-time monitoring: Catching things as they happen
      • Batch monitoring: Scheduled check-ups
    • Monitoring Deployed Graph Models
      • Performance monitoring: Is the model achieving its goal?
      • Structural monitoring: Is the graph itself changing?
      • Embedding monitoring: Are learned representations still meaningful?
    • Thresholds for Alerts
    • Visualization Tools
      • Real-time dashboards
      • Historical trends
      • Custom alerts
  • Feedback Loops in Graph Systems
    • User Feedback: Learning from Interaction
    • System Feedback: Optimizing Model Performance
    • Data Feedback: Adapting to a Changing World
    • Synergy Between Feedback Loops
    • Closed-Loop Systems: Automating Adaptation
      • What is a closed-loop system?
      • Key components of closed-loop graph systems
    • Adaptive Retraining Pipelines
      • Monitoring model health
      • Auto retraining triggering
      • Anomaly detection
    • Notification Systems
      • Alert design principles
      • Integration strategies
  • Summary
• 12. Future Trends: Graph Learning and LLMs
  • Introduction to Graph-Enhanced LLMs
    • LLMs and Their Transformative Impact
    • The Need for External Memory: The RAG Setup
    • Benefits of Graph-Enhanced RAG Setup
  • Retrieval Augmentation with Graph Integration
    • Challenges in Traditional Retrieval-Augmented Generation
    • Graph Integration
  • GraphRAG Methodology and Pipeline
    • Indexing Process
      • Text segmentation
      • Entity, relationship, and claim extraction
      • Knowledge graph construction
      • Hierarchical clustering and community formation
      • Embedding generation for graph nodes
      • Finalizing the index
    • Querying Process
      • Query parsing and understanding
      • Global search via community summaries
      • Localized entity search and graph traversal
      • Contextual augmentation for the LLM
      • Answer generation with evidence provenance
    • Enhanced Capabilities and Whole-Dataset Reasoning
    • Implementation Considerations
    • Baseline RAG Versus GraphRAG
  • RAG with Knowledge Graphs for Customer Service Question Answering
  • Summary
• Index

pokaż cały spis treści

ukryj recenzje