Dynamic Regret in Online Convex Optimization with Indicator Switching Costs

Oct 1, 2026·
Naram Mhaisen
Naram Mhaisen
,
George Iosifidis
Abstract
We study dynamic regret in online convex optimization with an indicator switching cost, a fixed penalty incurred whenever two consecutive decisions differ. We propose a meta-learning framework that is minimax-optimal, up to logarithmic factors, for tracking piecewise-constant comparators, without prior knowledge of their number of switches or path length.
Type
Publication
To appear in Advances in Neural Information Processing Systems (NeurIPS), 2026
publications
Online Learning Dynamic Regret Switching Costs
Naram Mhaisen
Authors
Naram Mhaisen
Postdoctoral Researcher

Naram is a Postdoctoral Researcher with the Networked Systems Group at the Faculty of Electrical Engineering, Mathematics, and Computer Science, TU Delft, The Netherlands. His research focuses on online learning techniques with applications in resource management across communication systems.

He received the B.Sc. degree in Computer Engineering from Qatar University in 2017 with the Order of Excellence award, and the M.Sc. degree with Distinction in Computing from Qatar University in 2020, supervised by Prof. Mohsen Guizani. He completed his Ph.D. at TU Delft in 2025, supervised by Dr. George Iosifidis, with a thesis on Optimistic Learning with Applications to Caching Networks.

His work has been recognized by the Microsoft Imagine Cup award (2017) and a NEC Europe Research Fellowship (2022).