Stochastic Regret Guarantees for Online Zeroth- and First-Order Bilevel Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Nazari, Parvin, Hou, Bojian, Tarzanagh, Davoud Ataee, Shen, Li, Michailidis, George
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916026722025472
author Nazari, Parvin
Hou, Bojian
Tarzanagh, Davoud Ataee
Shen, Li
Michailidis, George
author_facet Nazari, Parvin
Hou, Bojian
Tarzanagh, Davoud Ataee
Shen, Li
Michailidis, George
contents Online bilevel optimization (OBO) is a powerful framework for machine learning problems where both outer and inner objectives evolve over time, requiring dynamic updates. Current OBO approaches rely on deterministic \textit{window-smoothed} regret minimization, which may not accurately reflect system performance when functions change rapidly. In this work, we introduce a novel search direction and show that both first- and zeroth-order (ZO) stochastic OBO algorithms leveraging this direction achieve sublinear {stochastic bilevel regret without window smoothing}. Beyond these guarantees, our framework enhances efficiency by: (i) reducing oracle dependence in hypergradient estimation, (ii) updating inner and outer variables alongside the linear system solution, and (iii) employing ZO-based estimation of Hessians, Jacobians, and gradients. Experiments on online parametric loss tuning and black-box adversarial attacks validate our approach.
format Preprint
id arxiv_https___arxiv_org_abs_2511_01126
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Stochastic Regret Guarantees for Online Zeroth- and First-Order Bilevel Optimization
Nazari, Parvin
Hou, Bojian
Tarzanagh, Davoud Ataee
Shen, Li
Michailidis, George
Machine Learning
Numerical Analysis
Optimization and Control
Statistics Theory
Online bilevel optimization (OBO) is a powerful framework for machine learning problems where both outer and inner objectives evolve over time, requiring dynamic updates. Current OBO approaches rely on deterministic \textit{window-smoothed} regret minimization, which may not accurately reflect system performance when functions change rapidly. In this work, we introduce a novel search direction and show that both first- and zeroth-order (ZO) stochastic OBO algorithms leveraging this direction achieve sublinear {stochastic bilevel regret without window smoothing}. Beyond these guarantees, our framework enhances efficiency by: (i) reducing oracle dependence in hypergradient estimation, (ii) updating inner and outer variables alongside the linear system solution, and (iii) employing ZO-based estimation of Hessians, Jacobians, and gradients. Experiments on online parametric loss tuning and black-box adversarial attacks validate our approach.
title Stochastic Regret Guarantees for Online Zeroth- and First-Order Bilevel Optimization
topic Machine Learning
Numerical Analysis
Optimization and Control
Statistics Theory
url https://arxiv.org/abs/2511.01126