Score-based diffusion priors for multi-target detection

Alon Zabatani*, Shay Kreymer, Tamir Bendory

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Multi-target detection (MTD) is the problem of estimating an image from a large, noisy measurement that contains randomly translated and rotated copies of the image. Motivated by the single-particle cryo-electron microscopy technology, we design data-driven diffusion priors for the MTD problem, derived from score-based stochastic differential equations models. We then integrate the prior into the approximate expectation-maximization algorithm. In particular, our method alternates between an expectation step that approximates the expected log-likelihood and a maximization step that balances the approximated log-likelihood with the learned log-prior. We show on two datasets that adding the data-driven prior substantially reduces the estimation error, in particular in high noise regimes.

Original languageEnglish
Title of host publication2024 58th Annual Conference on Information Sciences and Systems, CISS 2024
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350369298
DOIs
StatePublished - 2024
Event58th Annual Conference on Information Sciences and Systems, CISS 2024 - Princeton, United States
Duration: 13 Mar 202415 Mar 2024

Publication series

Name2024 58th Annual Conference on Information Sciences and Systems, CISS 2024

Conference

Conference58th Annual Conference on Information Sciences and Systems, CISS 2024
Country/TerritoryUnited States
CityPrinceton
Period13/03/2415/03/24

Funding

FundersFunder number
United States-Israel Binational Science Foundation2020159
NSF-BSF2019752
Israel Science Foundation1924/21

    Keywords

    • Diffusion models
    • cryo-EM
    • expectation-maximization
    • multi-target detection
    • score-SDE

    Fingerprint

    Dive into the research topics of 'Score-based diffusion priors for multi-target detection'. Together they form a unique fingerprint.

    Cite this