Skip navigation
Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp01m039k765c
Full metadata record
DC FieldValueLanguage
dc.contributor.advisorSly, Allan-
dc.contributor.advisorAbbe, Emmanuel-
dc.contributor.authorBoix, Enric-
dc.date.accessioned2018-08-15T17:10:27Z-
dc.date.available2018-08-15T17:10:27Z-
dc.date.created2018-05-12-
dc.date.issued2018-08-15-
dc.identifier.urihttp://arks.princeton.edu/ark:/88435/dsp01m039k765c-
dc.description.abstractWe derive information-theoretic bounds for reconstruction in \(\mathbb{Z}/2\mathbb{Z}\) synchronization. Specifically, given a graph \(G\) whose vertices are labelled with i.i.d. Rademacher-\(1/2\) variables \(X_v\), and whose edges \((u,v)\) are labelled with outputs \(Y_{uv}\) of channels on \(X_u \cdot X_v\), we upper-bound the information that the edge labels give about the vertex labels. Our bounds relate the information given by \((X_u,Y)\) about \(X_v\) to the connection probability between \(u\) and \(v\) in a suitable bond percolation on \(G\). The proof is a simple interpolation argument. As applications of our bound, we re-derive known thresholds for impossibility of reconstruction in Broadcasting on Trees [EKPS00], for impossibility of recovery in the Spiked Gaussian Wigner Model [DAM15], and for impossibility of clustering in the Censored Block Model [LMX15]. Our bound also improves on the known threshold [AMM+17] for the impossibility of Grid Synchronization in the case of binary vertex labels.en_US
dc.format.mimetypeapplication/pdf-
dc.language.isoenen_US
dc.titleAn Information-Percolation Bound for \(\mathbb{Z}/2\mathbb{Z}\) Synchronizationen_US
dc.typePrinceton University Senior Theses-
pu.date.classyear2018en_US
pu.departmentMathematicsen_US
pu.pdf.coverpageSeniorThesisCoverPage-
pu.contributor.authorid960832026-
pu.certificateApplications of Computing Programen_US
Appears in Collections:Mathematics, 1934-2020

Files in This Item:
File Description SizeFormat 
BOIX-ENRIC-THESIS.pdf333.99 kBAdobe PDF    Request a copy


Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.