Alignment refers to the problem of establishing explicit correspondences between mathematical objects, such as structured or multimodal data, probability distributions, latent spaces, manifolds, and neural network models.

 

The TAGADA team proposes to work on the mathematical and foundational aspects of alignment, with the primary objective of designing methods and algorithms endowed with theoretical guaranties, as well as developing toolboxes for geometric alignment in machine learning. By formulating principled, geometry-aware, and data-efficient approaches to alignment, we focus on problem settings where insight, structure, and theory matter more than brute-force computation.

 

Our objective is to produce original, reusable, and scientifically well-founded methodologies that complement —rather than replicate— resource-intensive strategies, and that ultimately benefit the broader ML community.