Reset Method based on the Theory of Manifold Optimization on Real Manifolds

Weiping Liu, Jiajun Wang, He Li, Youfa Liu, Jingui Zou

Keywords: Manifold Optimization, Real Manifolds, Method, Deep Learning.

TL;DR: Based on the theory of real surface optimization, we propose a new optimization method, named the Reset Method.

Abstract: Manifold optimization is prominent in the fields of applied mathematics, statistics, machine learning, and in particular, deep learning. By leveraging the intrinsic geometric properties of manifolds, constrained optimization problems can be transformed into unconstrained optimization problems on certain manifolds. An innovative method, Reset Method, is introduced that combines manifold optimization and standard methods (SGD, Adam and AdamW), aiming to enhance the improvement of precision. The efficacy of our proposed method is corroborated by extensive deep learning experiments, providing visible higher precision.

Primary Area: optimization

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1161