An Active Learning Method Based on Variational Autoencoder and DBSCAN Clustering

Algorithm 1

Variational autoencoder for active learning.

Input: labeled pool , unlabeled pool , and initialized model
Input: hyperparameters: epochs and step size
(1)	for e = 1 to epochs do
(2)	Sample
(3)	Sample
(4)	Compute the reconstruction loss of VAE by using equation (1)
(5)	Compute the regularization loss of VAE by using equation (2)
(6)
(7)	Update VAE by descending stochastic gradients:
(8)	end for
(9)	return trained .