Patrick Perrot
Patrick Perrot is a French Gendarmerie General Officer who is working as the national AI coordinator for his institution. He has combined operational and scientific activities in order to fight against crime. He has a PhD in artificial intelligence and his main research interests regard biometrics, data mining, pattern recognition and public safety. He is also involved in different universities and an author of different publications in his fields of interest. He has been awarded with the ENFSI Referenced Best Article Award by the European Academy of Forensic Sciences.
Sessions
The emergence of Generative adversarial networks (GANs) has created new criminal opportunities through the generation of fake images, videos or texts, but also new possibilities to fight against crime and discrimination. This is what we call the double face of the use of GANs : Dr Jekyll and Mr Hyde. This study presents the two faces of the GAN's applications. A GAN is a generative model is based on two neural networks models : a generator and a discriminator and the combination is used to create new pausible examples. The two models are set up in a contest where the first one (generator) seeks to fool the second one (discriminator) model, and in the same time the discriminator is provided with both examples of real and generated samples. At the end, the generator is able to create new models on demand. In one hand GANs appears as a major breakthrough in the field of forgery and the potential for bad is there. It becomes easy to generate false news articles flooded almost all social media platforms, and develop a real propaganda. So, GAN's and their using in the case of deepfakes appears as a great challenge for LEA to fight against complotism, propaganda or cyber attacks on the balance of database. In another way GANs can provide relevant methods to authenticate documents, to generate photographs of faces with different ages from younger to older, in the case of child missing for instance, or to re-balance database in order to fight against discrimination and so on. Our work is on different cases using GANs and trying to understand the different faces of this new technique. We have for instance decomposed the principe of deepfake videos construction, developed a database to evaluate and to compensate bias to fight against discrimination and built a system able to detect fake images based on many GAN's methods.