Machine learning pour la compression d'image et de vidéo

Image and video compression continues to be a relevant scientific and technological concern with the continuous growth of video streaming.Recent statistics indicate that streaming accounts for more than 75% of the internet traffic. In this talk, I will first briefly recall the principles of classical compression methods, their current limits and the standardization landscape. Then, I will discuss how machine learning brings new methods and capabilities for compression. In particular, ML offers possibilities for semantic compression, as well as new paradigms based on implicit representations.