An Active Learning Method Based on Variational Autoencoder and DBSCAN Clustering

Algorithm 2

Sampling strategy by DBSCAN in the proposed model.

Input: , , query batch size: , and parameters: Eps and MinPts, N: total budget, and n: the amount of miniquery
(1)	Sample
(2)	Sample from the underlying distribution by using equation (4)
(3)	Sample as P randomly from and shuffle,
(4)	Cluster by adjusting Eps and MinPts
(5)	Remove noise
(6)	Sample all density-reachable unlabeled set C in all the clusters
(7)	fordo
(8)	Sample the needed amount of from C randomly, and find the corresponding original high-dimensional samples
(9)
(10)
(11)
(12)	end for
Output: ,