Detecting Insider Threat from Behavioral Logs Based on Ensemble and Self-Supervised Learning

Algorithm 2

Training algorithm for combined model.

Input: Set to legitimate sessions ; Set of legitimate sessions; T: Number of subdetectors; iter:Iterations number; m:Batch size; :learning rate of Detectors and User Recognizer; ID:User ID of input session
Output: Combined model and
(1)	initialize: for do rand_init
(2)	fordo
(3)	fordo
(4)
(5)	Over-Bootstrap
(6)	fordo
(7)
(8)
(9)
(10)	end for
(11)
(12)
(13)	end for
(14)	end for
(15)