Do I need to be Dr. Frankenstein to create real-ish synthetic data?
summary
Defines three synthesis methods — from real data via generative models (GANs), from analyst or simulation assumptions, and hybrids — and gives concrete metrics: Hellinger distance for univariate fits, correlation heatmaps and multivariate “make-each-feature-the-target” classification, and propensity score distinguishability via a binary classifier. Demonstrations use scikit‑learn’s make_classification, Nvidia graphics rendering for images, and libraries named pi db gen faker data synthesizer senpai and nike's time series generator; a Cambridge pilot uses synthetic data to validate a graph DB and then collects real data after promising results, and mixed real+synthetic training achieves low RMSE.
AI-written summary · about
chapters (11)
- 0:00Start
- 0:43About Myself
- 1:53Types of Synthetic Datasets
- 5:58Which one is for you?
- 7:46Why synthetic datasets?
- 10:54Some domain-specific use cases
- 15:54How to measure the realness?
- 16:16Comparing Bivariate Statistics
- 18:07Comparing Multivariate Statistics
- 18:44Distinguishability
- 25:27Resources
related talks
21mCrafting Flight Simulations: The Art of Synthetic Data Generation
01h 00mThe Synthetic Data Advantage: How It Helps You to Get Both AI Privacy & Explainability
14mLearning the Ropes of Synthetic Data
20mSynthetic Data for Privacy, AI Fairness and Democratizing Data Access
17mDon't Stop at Data Democratization!
54mOpen Data vs. Privacy - How Synthetic Data Can Democratize AI Innovation
28mInsights, Not Secrets: How Synthetic Data Empowers Privacy
16mCompute Engine Testing with Synthetic Data Generation