Scientists of the Krakow University of Technology (PK) are working on a deepfake detector in video, audio and images. They are planning to use modern techniques based on neural networks and deep learning methods to fight deepfakes.

Works on the project “Detection of manipulated audio-video content to protect against the spread of deepfake messages” are carried out under the 1st “Infostrateg” competition of the National Center for Research and Development. The Krakow team received funding of nearly PLN 4 million.

The project manager, Dr Eng. Michał Bereta, from the Faculty of Computer Science and Telecommunications of the Krakow University of Technology, assumes that the same or related machine learning methods that are used to produce deepfake content can be used to detect manipulated content.

As a final result, his team is going to create an automatic system for verifying the content of the image/sound / video to be used by media employees, editorial offices, PR agencies and other organizations protecting the image of their customers and products, Internet and social portals, the police and other security services, non-governmental organizations and everyone interested in checking the credibility of the content.

Scientists from the Krakow University of Technology are predicting that similarly to a constant arms race between the authors of computer viruses and the authors of antivirus programs, a comparable situation will take place between the creators of deepfake materials and researchers of neural networks and evolutionary computing who will try to develop the best methods of their detection.

Prof. Michał Bereta believes that the appearance of high-profile media attempts to use fake materials for the political struggle, defamation or fraud makes his project even more desirable by the market.

Adrian Andrzejewski