Abstract: | In this talk, we will discuss a special class of generative models, called diffusion processes, and present their applications as well as challenges that the scientific community is currently addressing.
We will first provide some background for the generative modelling task, and introduce the basic tools that are required to understand diffusion processes. We will then quickly overview the general topic of energy-based modelling, give some example applications, and focus on score-based generative models. Our goal is to provide an overview of the theory behind score-based models, but also to demystify them and suggest a practical interpretation of these methods. We will conclude the talk with some challenges we are currently addressing, including going beyond image data, multi-modality, and methodological advances that are required to improve the practical application of such exciting new methods |
Bio: | Pietro Michiardi received his M.S. in Electrical Engineering from Politecnico di Torino and his M.S. in Computer Science from EURECOM. Pietro received his Ph.D. in Computer Science from Telecom ParisTech, and his HDR (Habilitation) from UCA.
Today, Pietro is a Full Professor and head of the Data Science Department at EURECOM. In his work, Pietro focuses on scalable machine learning algorithms, Bayesian modelling and sampling, generative modelling and representation learning. Pietro is interested in developing a theoretical understanding of methods that learn from data, in methodological aspects of computationally efficient inference approaches, and their application to industrial domains such as the automotive, telecommunication and financial industries.
In the past, Pietro worked on a wide range of research topics, including: computer networks and their performance evaluation, applied cryptography, applied game theory, distributed systems, and data management systems. |
