000002324 001__ 2324
000002324 005__ 20210401130355.0
000002324 037__ $$aBELLE2-TALK-CONF-2021-016
000002324 041__ $$aeng
000002324 088__ $$aBELLE2-TALK-DRAFT-2021-027
000002324 100__ $$aStephanie Käs
000002324 245__ $$aBelle II pixeldetector cluster analyses using neural network algorithms
000002324 260__ $$aDPG Spring Meeting 2021$$c2021-03-09
000002324 300__ $$amult. p
000002324 520__ $$aThe Belle II DEPFET pixeldetector is operating since 2019, presently with 4 M pixels and trigger rates up to 5 kHz. The pixeldetector has the unique ability to detect exotic highly ionizing particles such as antideuterons or stable tetraquarks which due to their high energy loss do not reach the outer sub-detectors, and thus generate no reconstructable track. In order to identify these highly ionizing particles, multivariate analyses of pixeldetector clusters is performed. The multidimensional input space consists of variables such as single pixel signals, cluster observables, or Zernicke moments. We present results for cluster classification using different neural network algorithms: multilayer perceptrons, Convolutional networks, Kohonen-type networks (often denoted as self-organizing maps) and Hopfield-type networks (often denoted as associate memories). Data preprocessing by Principal Components analysis and possible implementation on an FPGA for online reconstruction are discussed as well.
000002324 700__ $$aJens Sören Lange
000002324 700__ $$aKatharina Dort
000002324 700__ $$aMarvin Peter
000002324 700__ $$aIrina Heinz
000002324 700__ $$aJohannes Bilk
000002324 700__ $$aPeter Lehnhardt
000002324 700__ $$aJohannes Budak
000002324 700__ $$aFalk Zorn
000002324 8560_ $$fsoerenlange@yahoo.com
000002324 8564_ $$uhttps://docs.belle2.org/record/2324/files/BELLE2-TALK-CONF-2021-016.pdf