State of the art-Preserving Privacy in Machine Learning and Data Protection

This state of the art will be led by Tom Sander, Chercheur, META AI (FAIR)

Deep Learning models are known for their capacity to memorize training data, a characteristic that can lead to significant privacy concerns.

In this talk, we will delve into the implications of this memorization and explore strategies to mitigate its effects. We will discuss practical methods to reduce unintended memorization and their results, providing insights into how these techniques can be deployed in real-world scenarios. We will also turn our attention to the concept of watermarking, a powerful tool that considerably helps in the authentication of AI-generated content. We will discuss how watermarking can also help confirm that a model was trained on a specific dataset, a capability that is particularly useful in identifying copyright infringement and detecting instances of model stealing or imitation.

This event is reserved for our members, but we still have a few places available for those who would like to discover the club. Don't hesitate to register: You'll be put on a waiting list and we'll confirm your participation 3 days before the event.

2404-Sander-PreservingPrivacy

Host or Publisher datacraft