-
heterogeneity, i.e., the fact that clients' local datasets are in general drawn from different distributions. Statistical heterogeneity for example slows down the convergence of FL algorithms [5]. In this thesis
-
approaches (e.g., GANs [2] or Plug& Play [3]). A different and increasingly popular class of methods producing outstanding results in many applied fields is based on the use of modern generative learning
-
fields for several applications in the field of computer vision and inverse problem [SLX+21]. As far as the modeling of data term between distributions is concerned, one idea would be also to follow
-
Virtual laboratory to predict the ability of a fluctuating biomass to satisfy a material use-VARIOUS
and organic waste) of which the properties are different and subject to seasonal and climatic variations. For the bioeconomy, it is therefore necessary to sort biomass sources according to the target
-
the following two scientific challenges. The first scientific challenge to address is how to effectively fuse the latent space of LiDAR-based models with VLM. This is challenging due to the difference between
-
, copyrighted, or biased. By studying brain data recordings and building computational models that mimic real populations of neurons, the project aims to uncover active unlearning: how the brain learns