Optimizing the Edge: Bringing Statistical Approaches to Decentralized Learning
-
Date: 17 OCTOBER 2025 from 17:30 to 19:00
-
Event location: Aula III via Francesco Selmi 2 - Bologna, Piano Terra. - In presence and online event
-
Type: Lectures
Speaker
ISA Visiting Fellow -Gesulado Scutari
Full Professor Purdue University
ATTENDANCE
-
in presence - book your seat
Book your seat within October 17, 12 p.m. The places will be assigned on “first come first served” basis.
The accessibility of the building is barrier-free, with pathway from the side entrance with building slide to the classroom located on the ground floor. Also available to persons with disabilities is a single-seat anthropometric bench with variable elevation and tilt positioned near the desk.
Over the past decade, Machine Learning and Artificial Intelligence (AI) have achieved significant breakthroughs, driven by data and computational power growth. These advancements have impacted fields like engineering, finance, medicine, and social media. Traditionally, these developments have adhered to a centralized paradigm, with data and decisionmaking concentrated in a few locations, such as data centers. However, the rise of mobile computing and the Internet of Things is rapidly shifting this paradigm, with data generated at the network edge now surpassing global data center traffic. This shift has led to the emergence of the so-called `edge intelligence’, focusing on moving computing and AI tasks from the network core to its edges. Decentralizing optimization and learning poses challenges at the intersection of computational and statistical sciences: maintaining analytics quality and trustworthiness despite limited resources at the edge. Understanding these dynamics remains underdeveloped. This is because most decentralized algorithms have been designed with an optimization focus, often neglecting statistical principles. This talk discusses some vignettes from highdimensional statistical inference, proposing new analyses and designs that bring statistical thinking to decentralized optimization, enhancing performance and reliability in high-dimensional, decentralized environments.