PEReN Lab

PEReN Lab is a window on the center’s activities from a more technical angle than usual. Through short, targeted articles, PEReN shares its state-of-the-art expertise on technologies deployed in its projects.

Generated and original content: an example of an approach for identifying their similarity in a dataset of artistic images

The rise of generative AI has led to the usage of millions of original images to generate new ones. In this context, identifying which specific original works were used to produce a particular artificial image is not only very costly in terms of computational power but also uncertain. PEReN explores an alternative approach in this prototype, based on the nearest neighbors search method, which allows for a low-cost identification of the images in the training set that are most similar to a generated image. Although imperfect, this approach provides a method for objective comparisons, serving as a basis for subsequent discussion.

Published on May 16, 2025

Detecting artificial content on social networks: a tool for knowing which detectors to trust

Generated content proliferates online, and the ability to detect it is now a major challenge to guarantee trust in AI and in information integrity. But it’s not an easy task! PEReN and Viginum set out to evaluate the performance and robustness to typical transformations observed on social networks of a set of detection models. Let’s dive in this open-source software.

Published on Feb 11, 2025

What are the risks and limits of generating convincing deepfakes?

The dazzling innovations in AI now make it possible for anyone to create deepfakes. But how easy is it to produce realistic images using the technologies available? In partnership with the CNIL, PEReN has carried out the following study.

Published on Dec 03, 2024

The French mobility orientation law on transport (2019) : an initial analysis of the statistical indicators published by digital labour platforms in France

Subject to an obligation of transparency, digital labour platforms have been publishing indicators on the activity of workers on delivery platforms or ride-hailing services since 2022. PEReN has studied these new data and offers a first unpublished critical analysis.

Published on Apr 25, 2024