| Zugriffsnummer | 52816 |
| Dokumenttyp | Zeitschriftenartikel |
| Peer Review | mit Peer Review |
| Sprache | Englisch |
| Titel | Mixed noise and posterior estimation with conditional deepGEM |
| Autor(in); Institution |
Hagemann, Paul; Institute of Mathematics, TU Berlin, Berlin, GERMANY
Hertrich, Johannes; Department of Computer Science, University College London, London, UK
Casfor Zapata, Maren; 8.4, Mathematische Modellierung und Datenanalyse, PTB-Berlin; 7.1, Radiometrie mit Synchrotronstrahlung, PTB-Berlin
Steidl, Gabriele; Institute of Mathematics, TU Berlin, Berlin, GERMANY
|
| Quelle/Jahr | Machine Learning: Science and Technology:(2024), 5, 035001-1 - 035001-17 |
| Artikelnummer | 035001 |
| Availability | [online only] |
| ISSN | 2632-2153 (online) |
| DOI | |
| Verlag | Bristol: IOP Publishing |
| Freie Schlagworte | normalizing flow ; inverse problem ; posterior ; expectation maximization |
| Zusammenfassung | We develop an algorithm for jointly estimating the posterior and the noise parameters in Bayesian inverse problems, which is motivated by indirect measurements and applications from nanometrology with a mixed noise model. We propose to solve the problem by an expectation maximization (EM) algorithm. Based on the current noise parameters, we learn in the E-step a conditional normalizing flow that approximates the posterior. In the M-step, we propose to find the noise parameter updates again by an EM algorithm, which has analytical formulas. We compare the training of the conditional normalizing flow with the forward and reverse Kullback–Leibler divergence, and show that our model is able to incorporate information from many measurements, unlike previous approaches. |
| Kostenfreier Zugang | Open Access Hybrid |
| Rechteinformation | CC BY 4.0 ; Creative Commons Attribution 4.0 License |
| Themenbereich der Metrologie | Metrologie in der Medizin |
| Forschungsprojekt | EMPIR ATMOC |
| Förderinformationen (1) |
Förderername: European Commission (EC)
Förderer ID: 0000 0001 2242 8989 Förderer ID Typ: ISNI Förderprogramm: EMPIR 2020 Industry Titel der Förderung: 20IND04: ATMOC: Traceable metrology of soft X-ray to IR optical constants and nanofilms for advanced manufacturing Förderungsnummer: 20IND04 |