Facebook Can Search Deepfake Video Sources

Facebook Can Search Deepfake Video Sources

Deepfakes are not a big problem at Facebook yet, but they are constantly looking for ways to deal with this technology.

In his research with academics from Michigan State University (MSU), Facebook found a way to reverse engineer deepfakes. That is by analyzing the AI-made image.

The analysis was carried out on the characteristics of the machine learning model used to create deepfake videos. From there, Facebook can help trace the source of the deepfake on various social media.

Why is this important? Because deepfakes often contain hoaxes and fake porn videos. But unfortunately, the method developed by Facebook is still in its early stages and is not ready for use to the public

In previous research on deepfakes, the AI ​​model used to create these deepfakes can already be detected. However, in the MSU research led by Vishal Asnani, the technology was continued with the ability to identify 'styles' that were not known before.

These 'styles' are known as hyperparameters, which must be reset for each machine learning model, just like a machine. But in the end they left a unique 'fingerprint' in the resulting image creating a deepfake. This fingerprint can then be used to identify the source of the deepfake.

This hyperparameter is very important according to the head of research from Facebook named Tal Hassner. The reason, according to him, is that deepfake software is very easy to customize, which makes it easy for perpetrators to cover up their digital footprints when traced by the authorities.

"Let's say a criminal creates a bunch of different deepfakes and uploads them on different platforms for different users. If this is an AI model that hasn't been used before, then there's not much we can do about it. But now we can see and recognize that all the images are coming from of the same model," Hassner said

Location:

Share :