File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Article: Nearly optimal bounds for the global geometric landscape of phase retrieval

TitleNearly optimal bounds for the global geometric landscape of phase retrieval
Authors
Keywordsgeometric landscape
nonconvex optimization
phase retrieval
Issue Date1-Jul-2023
PublisherIOP Publishing
Citation
Inverse Problems, 2023, v. 39, n. 7 How to Cite?
Abstract

The phase retrieval problem is concerned with recovering an unknown signal x∈Cn from a set of magnitude-only measurements yj=|⟨aj,x⟩|,j=1,…,m. A natural least squares formulation can be used to solve this problem efficiently even with random initialization, despite its non-convexity of the loss function. One way to explain this surprising phenomenon is the benign geometric landscape: (1) all local minimizers are global; and (2) the objective function has a negative curvature around each saddle point and local maximizer. In this paper, we show that m=O(nlog⁡n) Gaussian random measurements are sufficient to guarantee the loss function of a commonly used estimator has such benign geometric landscape with high probability. This is a step toward answering the open problem given by Sun et al (2018 Found. Comput. Math.18 1131–98), in which the authors suggest that O(nlog⁡n) or even O(n) is enough to guarantee the favorable geometric property.


Persistent Identifierhttp://hdl.handle.net/10722/347926
ISSN
2023 Impact Factor: 2.0
2023 SCImago Journal Rankings: 1.185

 

DC FieldValueLanguage
dc.contributor.authorCai, Jian Feng-
dc.contributor.authorHuang, Meng-
dc.contributor.authorLi, Dong-
dc.contributor.authorWang, Yang-
dc.date.accessioned2024-10-03T00:30:32Z-
dc.date.available2024-10-03T00:30:32Z-
dc.date.issued2023-07-01-
dc.identifier.citationInverse Problems, 2023, v. 39, n. 7-
dc.identifier.issn0266-5611-
dc.identifier.urihttp://hdl.handle.net/10722/347926-
dc.description.abstract<p>The phase retrieval problem is concerned with recovering an unknown signal x∈Cn from a set of magnitude-only measurements yj=|⟨aj,x⟩|,j=1,…,m. A natural least squares formulation can be used to solve this problem efficiently even with random initialization, despite its non-convexity of the loss function. One way to explain this surprising phenomenon is the benign geometric landscape: (1) all local minimizers are global; and (2) the objective function has a negative curvature around each saddle point and local maximizer. In this paper, we show that m=O(nlog⁡n) Gaussian random measurements are sufficient to guarantee the loss function of a commonly used estimator has such benign geometric landscape with high probability. This is a step toward answering the open problem given by Sun <em>et al</em> (2018 <em>Found. Comput. Math.</em><strong>18</strong> 1131–98), in which the authors suggest that O(nlog⁡n) or even <em>O</em>(<em>n</em>) is enough to guarantee the favorable geometric property.</p>-
dc.languageeng-
dc.publisherIOP Publishing-
dc.relation.ispartofInverse Problems-
dc.subjectgeometric landscape-
dc.subjectnonconvex optimization-
dc.subjectphase retrieval-
dc.titleNearly optimal bounds for the global geometric landscape of phase retrieval-
dc.typeArticle-
dc.identifier.doi10.1088/1361-6420/acdab7-
dc.identifier.scopuseid_2-s2.0-85163732113-
dc.identifier.volume39-
dc.identifier.issue7-
dc.identifier.eissn1361-6420-
dc.identifier.issnl0266-5611-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats