Supervised Deep Learning in High Energy Phenomenology: a Mini Review *

Murat Abdughani,Jie Ren,Lei Wu,Jin-Min Yang,Jun Zhao

Fig. 18 (Color online) The structure of our MPNN designed for event graph classification. It has the functional layers shown as shadowed blocks, with $T$ pairs of message passing and state updating layers for automatic event feature extraction. The state vectors ${s}$, the message vectors ${m}$, the votes $y_i$ and the discrimination score $y$ are shown as black boxes. The colored arrows denote the application of node embedding function $f_e$, message passing function $f_m$, state update function $f_u$ and vote function $f_v$, respectively. The operators are given in gray circles, with $\oplus$ denoting vector concatenation, $\Sigma$, and $\Sigma/N$ being summation and average, respectively. This figure is taken from Ref. [13].