Synthetic Data for Machine Learning : Revolutionize Your Approach to Machine Learning with This Comprehensive Conceptual Guide

Synthetic Data for Machine Learning Revolutionize Your Approach to Machine Learning with This Comprehensive Conceptual Guide

Conquer data hurdles, supercharge your ML journey, and become a leader in your field with synthetic data generation techniques, best practices, and case studies Key Features Avoid common data issues by identifying and solving them using synthetic data-based solutions Master synthetic data generation...

Descripción completa

Detalles Bibliográficos
Autor principal: Kerim, Abdulrahman (-)
Formato: Libro electrónico
Idioma:Inglés
Publicado: Birmingham : Packt Publishing, Limited 2023.
Edición:1st ed
Materias:
Machine learning.
Computer vision.
Ver en Biblioteca Universitat Ramon Llull:https://discovery.url.edu/permalink/34CSUC_URL/1im36ta/alma991009781238006719
Tabla de Contenidos:
  • Cover
  • Title Page
  • Copyright and Credits
  • Dedications
  • Contributors
  • Table of Contents
  • Part 1: Real Data Issues, Limitations, and Challenges
  • Chapter 1: Machine Learning and the Need for Data
  • Technical requirements
  • Artificial intelligence, machine learning, and deep learning
  • Artificial intelligence (AI)
  • Machine learning (ML)
  • Deep learning (DL)
  • Why are ML and DL so powerful?
  • Feature engineering
  • Transfer across tasks
  • Training ML models
  • Collecting and annotating data
  • Designing and training an ML model
  • Validating and testing an ML model
  • Iterations in the ML development process
  • Summary
  • Chapter 2: Annotating Real Data
  • Annotating data for ML
  • Learning from data
  • Training your ML model
  • Testing your ML model
  • Issues with the annotation process
  • The annotation process is expensive
  • The annotation process is error-prone
  • The annotation process is biased
  • Optical flow and depth estimation
  • Ground truth generation for computer vision
  • Optical flow estimation
  • Depth estimation
  • Summary
  • Chapter 3: Privacy Issues in Real Data
  • Why is privacy an issue in ML?
  • ML task
  • Dataset size
  • Regulations
  • What exactly is the privacy problem in ML?
  • Copyright and intellectual property infringement
  • Privacy and reproducibility of experiments
  • Privacy issues and bias
  • Privacy-preserving ML
  • Approaches for privacy-preserving datasets
  • Approaches for privacy-preserving ML
  • Real data challenges and issues
  • Summary
  • Part 2: An Overview of Synthetic Data for Machine Learning
  • Chapter 4: An Introduction to Synthetic Data
  • Technical requirements
  • What is synthetic data?
  • Synthetic and real data
  • Data-centric and architecture-centric approaches in ML
  • History of synthetic data
  • Random number generators
  • Generative Adversarial Networks (GANs).
  • Synthetic data for privacy issues
  • Synthetic data in computer vision
  • Synthetic data and ethical considerations
  • Synthetic data types
  • Data augmentation
  • Geometric transformations
  • Noise injection
  • Text replacement, deletion, and injection
  • Summary
  • Chapter 5: Synthetic Data as a Solution
  • The main advantages of synthetic data
  • Unbiased
  • Diverse
  • Controllable
  • Scalable
  • Automatic data labeling
  • Annotation quality
  • Low cost
  • Solving privacy issues with synthetic data
  • Using synthetic data to solve time and efficiency issues
  • Synthetic data as a revolutionary solution for rare data
  • Synthetic data generation methods
  • Summary
  • Part 3: Synthetic Data Generation Approaches
  • Chapter 6: Leveraging Simulators and Rendering Engines to Generate Synthetic Data
  • Introduction to simulators and rendering engines
  • Simulators
  • Rendering and game engines
  • History and evolution of simulators and game engines
  • Generating synthetic data
  • Identify the task and ground truth to generate
  • Create the 3D virtual world in the game engine
  • Setting up the virtual camera
  • Adding noise and anomalies
  • Setting up the labeling pipeline
  • Generating the training data with the ground truth
  • Challenges and limitations
  • Realism
  • Diversity
  • Complexity
  • Looking at two case studies
  • AirSim
  • CARLA
  • Summary
  • Chapter 7: Exploring Generative Adversarial Networks
  • Technical requirements
  • What is a GAN?
  • Training a GAN
  • GAN training algorithm
  • Training loss
  • Challenges
  • Utilizing GANs to generate synthetic data
  • Hands-on GANs in practice
  • Variations of GANs
  • Conditional GAN (cGAN)
  • CycleGAN
  • Conditional Tabular GAN (CTGAN)
  • Wasserstein GAN (WGAN) and Wasserstein GAN with Gradient Penalty (WGAN-GP)
  • f-GAN
  • DragGAN
  • Summary
  • Chapter 8: Video Games as a Source of Synthetic Data.
  • The impact of the video game industry
  • Photorealism and the real-synthetic domain shift
  • Time, effort, and cost
  • Generating synthetic data using video games
  • Utilizing games for general data collection
  • Utilizing games for social studies
  • Utilizing simulation games for data generation
  • Challenges and limitations
  • Controllability
  • Game genres and limitations on synthetic data generation
  • Realism
  • Ethical issues
  • Intellectual property
  • Summary
  • Chapter 9: Exploring Diffusion Models for Synthetic Data
  • Technical requirements
  • An introduction to diffusion models
  • The training process of DMs
  • Applications of DMs
  • Diffusion models - the pros and cons
  • The pros of using DMs
  • The cons of using DMS
  • Hands-on diffusion models in practice
  • Context
  • Dataset
  • ML model
  • Training
  • Testing
  • Diffusion models - ethical issues
  • Copyright
  • Bias
  • Inappropriate content
  • Responsibility
  • Privacy
  • Fraud and identity theft
  • Summary
  • Part 4: Case Studies and Best Practices
  • Chapter 10: Case Study 1 - Computer Vision
  • Transforming industries - the power of computer vision
  • The four waves of the industrial revolution
  • Industry 4.0 and computer vision
  • Synthetic data and computer vision - examples from industry
  • Neurolabs using synthetic data in retail
  • Microsoft using synthetic data alone for face analysis
  • Synthesis AI using synthetic data for virtual try-on
  • Summary
  • Chapter 11: Case Study 2 - Natural Language Processing
  • A brief introduction to NLP
  • Applications of NLP in practice
  • The need for large-scale training datasets in NLP
  • Human language complexity
  • Contextual dependence
  • Generalization
  • Hands-on practical example with ChatGPT
  • Synthetic data as a solution for NLP problems
  • SYSTRAN Soft's use of synthetic data
  • Telefónica's use of synthetic data.
  • Clinical text mining utilizing synthetic data
  • The Alexa virtual assistant model
  • Summary
  • Chapter 12: Case Study 3 - Predictive Analytics
  • What is predictive analytics?
  • Applications of predictive analytics
  • Predictive analytics issues with real data
  • Partial and scarce training data
  • Bias
  • Cost
  • Case studies of utilizing synthetic data for predictive analytics
  • Provinzial and synthetic data
  • Healthcare benefits from synthetic data in predictive analytics
  • Amazon fraud transaction prediction using synthetic data
  • Summary
  • Chapter 13: Best Practices for Applying Synthetic Data
  • Unveiling the challenges of generating and utilizing synthetic data
  • Domain gap
  • Data representation
  • Privacy, security, and validation
  • Trust and credibility
  • Domain-specific issues limiting the usability of synthetic data
  • Healthcare
  • Finance
  • Autonomous cars
  • Best practices for the effective utilization of synthetic data
  • Summary
  • Part 5: Current Challenges and Future Perspectives
  • Chapter 14: Synthetic-to-Real Domain Adaptation
  • The domain gap problem in ML
  • Sensitivity to sensors' variations
  • Discrepancy in class and feature distributions
  • Concept drift
  • Approaches for synthetic-to-real domain adaptation
  • Domain randomization
  • Adversarial domain adaptation
  • Feature-based domain adaptation
  • Synthetic-to-real domain adaptation - issues and challenges
  • Unseen domain
  • Limited real data
  • Computational complexity
  • Synthetic data limitations
  • Multimodal data complexity
  • Summary
  • Chapter 15: Diversity Issues in Synthetic Data
  • The need for diverse data in ML
  • Transferability
  • Better problem modeling
  • Security
  • Process of debugging
  • Robustness to anomalies
  • Creativity
  • Inclusivity
  • Generating diverse synthetic datasets
  • Latent space variations.
  • Ensemble synthetic data generation
  • Diversity regularization
  • Incorporating external knowledge
  • Progressive training
  • Procedural content generation with game engines
  • Diversity issues in the synthetic data realm
  • Balancing diversity and realism
  • Privacy and confidentiality concerns
  • Validation and evaluation challenges
  • Summary
  • Chapter 16: Photorealism in Computer Vision
  • Synthetic data photorealism for computer vision
  • Feature extraction
  • Domain gap
  • Robustness
  • Benchmarking performance
  • Photorealism approaches
  • Physically Based Rendering (PBR)
  • Neural style transfer
  • Photorealism evaluation metrics
  • Structural Similarity Index Measure (SSIM)
  • Learned Perceptual Image Patch Similarity (LPIPS)
  • Expert evaluation
  • Challenges and limitations of photorealistic synthetic data
  • Creating hyper-realistic scenes
  • Resources versus photorealism trade-off
  • Summary
  • Chapter 17: Conclusion
  • Real data and its problems
  • Synthetic data as a solution
  • Real-world case studies
  • Challenges and limitations
  • Future perspectives
  • Summary
  • Index
  • Other Books You May Enjoy.

Ejemplares similares