Author: Ben Ghali, M.
Paper Title Page
THPAB201 A Machine Learning Technique for Dynamic Aperture Computation 4172
 
  • B. Dalena, M. Ben Ghali
    CEA-IRFU, Gif-sur-Yvette, France
  
  Cur­rently, dy­namic aper­ture cal­cu­la­tions of high-en­ergy hadron col­lid­ers are per­formed through com­puter sim­u­la­tions, which are both a re­source-heavy and time-costly processes. The aim of this study is to use a reser­voir com­put­ing ma­chine learn­ing model in order to achieve a faster ex­trap­o­la­tion of dy­namic aper­ture val­ues. A re­cur­rent echo-state net­work (ESN) ar­chi­tec­ture is used as a basis for this work. Re­cur­rent net­works are bet­ter fit­ted to ex­trap­o­la­tion tasks while the reser­voir echo-state struc­ture is com­pu­ta­tion­ally ef­fec­tive. Model train­ing and val­i­da­tion is con­ducted on a set of "seeds" cor­re­spond­ing to the sim­u­la­tion re­sults of dif­fer­ent ma­chine con­fig­u­ra­tions. Ad­just­ments in the model ar­chi­tec­ture, man­ual met­ric and data se­lec­tion, hy­per-pa­ra­me­ters tun­ing and the in­tro­duc­tion of new pa­ra­me­ters en­abled the model to re­li­ably achieve good per­for­mance on ex­am­in­ing test­ing sets.  
DOI • reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-THPAB201  
About • paper received ※ 14 May 2021       paper accepted ※ 22 July 2021       issue date ※ 02 September 2021  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)