第75回統計的機械学習セミナー / The 75th Statistical Machine Learning Seminar (Hybrid)

【Date & Time】
31 July, 2026 (Friday)　11:00 - 12:00
Admission Free, No Booking Necessary
【Place】
D313/314, 3rd floor, The Institute of Statistical Mathematics
Online :
Please register from the URL below to get a Zoom link:
https://forms.gle/MtnohSNnhszZhmcX8
【Speaker】
Andrea Paudice (Aarhus University)
【Title】
General Tail Bounds for Non-Smooth Stochastic Mirror Descent
【Abstract】
We study the problem of minimizing a convex, non-smooth Lipschitz function over a convex domain when only noisy stochastic subgradient estimates are available. We analyze the classical Stochastic Mirror Descent (SMD) algorithm and derive new tail bounds on its optimization error, for both the averaged and the last iterate. Our results extend existing analyses - traditionally limited to light-tailed, sub-Gaussian noise - to heavier-tailed noise distributions. We specialize our general bounds to two important families of noise: one with exponential tails and another with polynomial tails. Notably, our bounds for the averaged iterate reveal a distinct two-regime behaviour, highlighting new insights into the interplay between noise tails and convergence rates.
【Short Bio】
Andrea Paudice is a tenure-track Assistant Professor in Computer Science at Aarhus University. Previously, he held a joint postdoctoral position at Italian Institute of Technology and the University of Milan (Statale), where he obtained his PhD in Computer Science under the supervision of Nicolò Cesa-Bianchi. Before that, he spent approximately three years as a Research Fellow at Imperial College London. His research interests lie in the theory of machine learning, with a focus on stochastic optimization, generalization bounds, and the analysis of classical algorithms in non-standard settings.