Improved Stochastic Optimization of LogSumExp

Egor Gladin, Alexey Kroshnin, Jia-Jie Zhu, Pavel Dvurechensky · 2025

Read Improved Stochastic Optimization of LogSumExp by Egor Gladin, Alexey Kroshnin, Jia-Jie Zhu, Pavel Dvurechensky (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-2509-24894.txt.

Description

Egor Gladin and co-authors improve this 2025 stochastic optimization of a workhorse function, LogSumExp. The function sits inside transport problems and robust optimization, as the mirror image of a famous divergence measure. When it sums huge or endless numbers of terms, computing its slope exactly is the bottleneck. The team built an approximation that keeps the original's convexity and smoothness.

Their approximation comes from a safer variant of the divergence, designed to behave under sampling noise. Stochastic gradient methods can then attack it directly. Theory and experiments on robust optimization and continuous transport tasks showed gains over existing baselines. The work targets the many pipelines where this one function quietly dominates the compute bill.

Catalog record

CREATOREgor Gladin, Alexey Kroshnin, Jia-Jie Zhu, Pavel Dvurechensky
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-24894

Browse the full library · About the Tower of Babel