Synthetic Data May Be the Answer for Using AI in Biometrics

The AI Act will make the gathering and processing of personal data for biometric purposes even more complicated than it currently is. Using synthetic data for training might be the answer to this challenge.

There’s more to synthetic data for AI training than just robust datasets. Laws and regulations will make it only more difficult to work with real data, especially in biometrics. Successful work with synthetic data will be paramount in the future while also protecting privacy and preventing bias.

Using synthetic data in machine learning was one of the topics of the Eastern European Machine Learning Summer School organized by Google DeepMind in Kosice, Slovakia. Innovatrics image synthesis team leader Igor Janos explained the ways Innovatrics has been using the generated data for improving algorithms.

“There are of course obstacles to the successful synthesization of data. For example, you need large amounts of source data to train your generative model. However, the main reason you usually need synthetic data is that you don’t have enough real data. And that is the contradiction,” Igor Janos explains.

In Innovatrics, we use synthetic fingerprint fragments to improve algorithms that detect and identify latent fingerprints that can be found on a crime scene.

You can watch more in-depth analysis by our Image Synthesis Team Leader Igor Janos from his Eastern European Machine Learning Summer School EEML presentation here.