History of synthetic data

In this section, we will learn about the evolution of synthetic data. Basically, we can categorize the use of synthetic data into the following categories, which may not reflect the chronological order, as it is very hard to track the early uses of synthetic data for each category.

Random number generators

Random number generators are one of the simplest forms of synthetic data. Assume you are training an ML model to recognize faces. Let us say you have only a limited number of images. You can add, for example, random noise to the original images to create new synthetic ones. The implementation of random noise is possible through the utilization of random number generators. This will help the face recognizer ML model to learn how the person’s face changes under certain types of noise (see Figure 4.3).

Figure 4.3 – Utilizing random number generators to generate synthetic images

Next, we’ll learn about...