Synthetic Data

1.) It depends on the cost of acquiring real data. If you can easily obtain a large amount of real data, including labels, go for it. A good example for synthetic data is medical: you need experts – professors or doctors – which makes it extremely expensive to obtain labelled data at scale. There’s a strong case for synthetic data.

2.) In the past, synthetic data was generated only by computer graphics. Now the trend goes to AI-generated data, where generative models increasingly automate parts of these computer graphics pipelines.

Technische Hochschule Mannheim

www.hs-mannheim.de