Fixed Confidence and Fixed Tolerance Bi-level Optimization for Selecting the Best Optimized System
Yuhao Wang, Seong-Hee Kim, Enlu Zhou
TL;DR
This work formulates a fixed-confidence, fixed-tolerance (FCFT) bilevel optimization problem to select the best system from a finite set where the upper-level performance and lower-level decision variables interact but may have different objectives. It introduces a two-mode strategy: a direct two-step baseline and a scalable multi-stage pruning-optimization framework that progressively reduces the candidate set by pruning inferior systems and refining remaining ones with decreasing tolerances. The methodology combines fully sequential pruning steps with both non-asymptotic (SAGD) and asymptotic (SGD) optimization analyses to ensure statistical guarantees while improving sample efficiency. Theoretical results establish the framework’s validity and sample complexity, and numerical experiments on drug dosage optimization demonstrate substantial computational savings, especially when using asymptotic concentration bounds for the optimization step.
Abstract
In this paper, we study a fixed-confidence, fixed-tolerance formulation of a class of stochastic bi-level optimization problems, where the upper-level problem selects from a finite set of systems based on a performance metric, and the lower-level problem optimizes continuous decision variables for each system. Notably, the objective functions for the upper and lower levels can differ. This class of problems has a wide range of applications, including model selection, ranking and selection under input uncertainty, and optimal design. To address this, we propose a multi-stage Pruning-Optimization framework that alternates between comparing the performance of different systems (Pruning) and optimizing systems (Optimization). % In the Pruning stage, we design a sequential algorithm that identifies and eliminates inferior systems through systematic performance evaluations. In the Optimization stage, the goal is to solve for a near-optimal solution that meets specified confidence and tolerance requirements. This multi-stage framework is designed to enhance computational efficiency by pruning inferior systems with high tolerance early on, thereby avoiding unnecessary computational efforts. We demonstrate the effectiveness of the proposed algorithm through both theoretical analysis of statistical validity and sample complexity and numerical experiments.