New and future developments of geometric deep learning
Date: Wed 25 Mar 2020
Speaker: Emanuele Rodolà
Title: New and future developments of geometric deep learning
Abstract: In this talk I will present some new formulations of deep learning for non-Euclidean structured data such as graphs and manifolds, which are becoming increasingly important in a variety of fields including computer vision, graphics, biomedicine, and computational social sciences. I will start by presenting the general ideas underpinning this area of research, and present very recent results we obtained in the field of computational biology for the prediction of protein interactions. I will then move on to present some open challenges and future work I am currently pursuing with my group, showing how even stronger integration of geometric knowledge (either in the primal, or in the dual frequency domain) into deep learning pipelines holds great promise for tough problems in a wide range of applicative domains.