Detecting Unknown Threat Based on Continuous-Time Dynamic Heterogeneous Graph Network

Table 5

Detection result on different number of attention heads in a CDHGN layer.

#Multihead (LayerNorm)	Dataset division	#Train	#Val	#Test	AUC (test)
2-head	0.22 : 0.04 : 0.74	152	28	511	0.9400
8-head	0.22 : 0.04 : 0.74	152	28	511	0.9348
16-head	0.22 : 0.04 : 0.74	152	28	511	0.9400
32-head	0.22 : 0.04 : 0.74	152	28	511	0.9316