Make Optimization Once and for All with Fine-grained Guidance

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shi, Mingjia, Lin, Ruihan, Chen, Xuxi, Zhou, Yuhao, Ding, Zezhen, Li, Pingzhi, Wang, Tong, Wang, Kai, Wang, Zhangyang, Zhang, Jiheng, Chen, Tianlong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913736225193984
author Shi, Mingjia
Lin, Ruihan
Chen, Xuxi
Zhou, Yuhao
Ding, Zezhen
Li, Pingzhi
Wang, Tong
Wang, Kai
Wang, Zhangyang
Zhang, Jiheng
Chen, Tianlong
author_facet Shi, Mingjia
Lin, Ruihan
Chen, Xuxi
Zhou, Yuhao
Ding, Zezhen
Li, Pingzhi
Wang, Tong
Wang, Kai
Wang, Zhangyang
Zhang, Jiheng
Chen, Tianlong
contents Learning to Optimize (L2O) enhances optimization efficiency with integrated neural networks. L2O paradigms achieve great outcomes, e.g., refitting optimizer, generating unseen solutions iteratively or directly. However, conventional L2O methods require intricate design and rely on specific optimization processes, limiting scalability and generalization. Our analyses explore general framework for learning optimization, called Diff-L2O, focusing on augmenting sampled solutions from a wider view rather than local updates in real optimization process only. Meanwhile, we give the related generalization bound, showing that the sample diversity of Diff-L2O brings better performance. This bound can be simply applied to other fields, discussing diversity, mean-variance, and different tasks. Diff-L2O's strong compatibility is empirically verified with only minute-level training, comparing with other hour-levels.
format Preprint
id arxiv_https___arxiv_org_abs_2503_11462
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Make Optimization Once and for All with Fine-grained Guidance
Shi, Mingjia
Lin, Ruihan
Chen, Xuxi
Zhou, Yuhao
Ding, Zezhen
Li, Pingzhi
Wang, Tong
Wang, Kai
Wang, Zhangyang
Zhang, Jiheng
Chen, Tianlong
Machine Learning
68Q32
I.2
Learning to Optimize (L2O) enhances optimization efficiency with integrated neural networks. L2O paradigms achieve great outcomes, e.g., refitting optimizer, generating unseen solutions iteratively or directly. However, conventional L2O methods require intricate design and rely on specific optimization processes, limiting scalability and generalization. Our analyses explore general framework for learning optimization, called Diff-L2O, focusing on augmenting sampled solutions from a wider view rather than local updates in real optimization process only. Meanwhile, we give the related generalization bound, showing that the sample diversity of Diff-L2O brings better performance. This bound can be simply applied to other fields, discussing diversity, mean-variance, and different tasks. Diff-L2O's strong compatibility is empirically verified with only minute-level training, comparing with other hour-levels.
title Make Optimization Once and for All with Fine-grained Guidance
topic Machine Learning
68Q32
I.2
url https://arxiv.org/abs/2503.11462