Introduction to Synthetic Data for Machine Learning
Chapter 1: Understanding Synthetic Data
What is Synthetic Data?
Why Generate Artificial Data?
Real Data vs. Synthetic Data
Common Terminology
Potential Benefits
General Limitations
Quiz for Chapter 1
Chapter 2: Basic Methods for Data Generation
The Idea of Data Generation Models
Generating Data from Statistical Distributions
Introduction to Rule-Based Systems
Generating Simple Numerical Data
Generating Simple Categorical Data
Hands-on Practical: Create Basic Synthetic Data
Quiz for Chapter 2
Chapter 3: Generating Synthetic Tabular Data
Understanding Tabular Data Structure
Row Sampling Techniques
Independent Column Value Generation
Preserving Basic Column Correlations
Introduction to Data Anonymization Concepts
Hands-on Practical: Generate a Synthetic Table
Quiz for Chapter 3
Chapter 4: Introduction to Synthetic Image Data
Why Synthetic Data for Images?
Basic Image Properties: Pixels and Color
Creating Images with Simple Shapes and Patterns
Applying Noise and Simple Augmentations
Introduction to Rendering Simple Scenes
Challenges in Realistic Image Generation
Hands-on Practical: Generate Simple Synthetic Images
Quiz for Chapter 4
Chapter 5: Evaluating Synthetic Data Quality
Importance of Evaluation
Visual Inspection Methods
Basic Statistical Comparisons
Comparing Data Distributions
Concept of Fidelity vs. Utility
Quiz for Chapter 5
Chapter 6: Tools and Libraries Overview
Role of Software in Data Generation
Libraries for Basic Data Manipulation (NumPy, Pandas)
Introduction to Faker Library
Libraries for Simple Image Manipulation (Pillow, Scikit-image)
Finding Generation Tools
Quiz for Chapter 6
Common Terminology
Was this section helpful?
- A Survey of Synthetic Data Generation Methods for Tabular Data, Viktoriia Borisov, Tobias Leemann, Jörg Huber, Marc Fischer, Tobias Böhm, Michele F. Sutter, and Stefan Feuerriegel, 2022 IEEE Transactions on Knowledge and Data Engineering (IEEE) DOI: 10.1109/TKDE.2022.3214876 - This comprehensive survey provides a detailed overview of synthetic data generation methods for tabular data, a primary focus of the course. It covers various approaches and discusses their strengths and weaknesses.
- What makes good synthetic data? Statistical and data utility metrics for generative models, Kyle M. O’Donoghue, Ryan D. Hays, Andrew A. Kress, and Daniel E. Casey, 2022 Journal of the American Medical Informatics Association, Vol. 29 (Oxford University Press) DOI: 10.1093/jamia/ocac014 - This paper specifically examines and categorizes various metrics for evaluating the quality of synthetic data, focusing on both statistical fidelity and practical utility for machine learning tasks.
- A Survey on Privacy-Preserving Synthetic Data Generation, Sangmin Kim, Tuan Anh Nguyen, and Yansong Li, 2021 ACM Computing Surveys, Vol. 54 (Association for Computing Machinery) DOI: 10.1145/3478546 - This survey reviews methods and considerations for generating synthetic data while preserving privacy, including discussions on techniques relevant to anonymization and advanced privacy mechanisms.
- Synthetic Data Generation for Machine Learning: A Practical Guide, Alistair Jordon, Paul van der Putten, Jinsung Yoon, Mihaela van der Schaar, and Richard Schmarzo, 2022 (IBM Research) - This practical guide from IBM Research offers a clear introduction to synthetic data generation for machine learning, explaining fundamental concepts and their applications.