Superquantile-Gibbs Relaxation for Minima-selection in Bi-Level Optimization

Saeed Masiha, Zebang Shen, Negar Kiyavash, Niao He · 2025

Read Superquantile-Gibbs Relaxation for Minima-selection in Bi-Level Optimization by Saeed Masiha, Zebang Shen, Negar Kiyavash, Niao He (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-05991.txt.

Introduction

Two-level problems, where one choice sets the stage for a second, grow hard when the inner problem has many best answers. Saeed Masiha and three co-authors tackle that case in 2025. Picking which inner answer counts can break the smoothness of the outer goal, which stalls standard tools.

The team works under a local rule that asks the inner goal to curve up near its best points, but only there. Under that rule, they show the outer goal changes in a smooth, bounded way, and the set of inner answers forms a neat curved sheet. They then build a softened method that relaxes the pick step, in the spirit of a classic physics tool. The work is aimed at tasks like tuning learning models that are themselves trained to a fit.

Catalog record

CREATORSaeed Masiha, Zebang Shen, Negar Kiyavash, Niao He
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE174 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-05991

Browse the full library · About the Tower of Babel