BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Harold Erbin: Machine learning for complete intersection Calabi-Ya
u manifolds
DTSTART;VALUE=DATE-TIME:20220202T151500Z
DTEND;VALUE=DATE-TIME:20220202T161500Z
DTSTAMP;VALUE=DATE-TIME:20230603T150000Z
UID:indico-event-7096@indico.math.cnrs.fr
DESCRIPTION:In this talk\, I will explain how to compute both Hodge number
s for complete intersection Calabi-Yau (CICY) 3- and 4-folds using machine
learning. I will first make a tour of various machine learning algorithms
and explain how exploratory data analysis can help in improving results.
Then\, I will describe a neural network inspired from the Google's Incepti
on model which\, for the 3-folds\, reaches nearly perfect accuracy for h11
using much fewer data and parameters compared to other approaches. I will
then describe an improved architecture to compute 3 out of 4 Hodge number
s for the 4-folds with more than 95% accuracy. I will conclude by describi
ng how more recent techniques could improve the computations of the remain
ing Hodge numbers and extract analytic information from the network.\n\nar
xiv: 2007.13379\, 2007.15706\, 2108.02221\n\nhttps://indico.math.cnrs.fr/e
vent/7096/
LOCATION:salle 318
URL:https://indico.math.cnrs.fr/event/7096/
END:VEVENT
END:VCALENDAR