{"id":"arxiv/2610.15047","url":"https://arxiv.org/abs/2610.15047","tags":["math.OC","math.NA"],"score":0,"title":"Multilevel Monte Carlo for Zeroth-Order Stochastic Bilevel Optimization with Inexact Lower Level Solves","domain":"math.OC","source":"arxiv","authors":["Neil K. Chada","Mohammad S. Salehi","Mathias Staudigl","Simon Weissmann"],"reviews":0,"subject":"math.OC","summary":"Variational regularization techniques are dominant in the field of mathematical imaging. The expressiveness of these methods heavily depends on the adopted regularization formulation which usually involves a number of parameters which have to be fine-tuned by the user. To circumvent the heavy computational costs associated with this parameter tuning, data-driven learning strategies have turned out to be a useful strategy. Selecting parameters by learning from data gives rise to a hierarchical (bilevel) optimization problem, which is in general computationally hard to solve. Our aim in this…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-10","reactions":{}}