Feature Dimensionality Reduction with Variational Autoencoders in Deep Bayesian Active Learning.

Col P. E., ERTEKİN BOLELLİ Ş.

29th Signal Processing and Communications Applications Conference (SIU 2021), İstanbul, Türkiye, 09 Haziran 2021 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu53274.2021.9477979
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

Data annotation for training of supervised learning algorithms has been a very costly procedure. The aim of deep active learning methodologies is to acquire the highest performance in supervised deep learning models by annotating as few data points as possible. As the feature space of data grows, the application of linear models in active learning settings has become insufficient. Therefore, Deep Bayesian Active Learning methodology which represents model uncertainty has been widely studied. In this paper, a study has been conducted in order to increase the performance of Deep Bayesian Active Learning method. Feature dimensionality reduction is performed on data set by using Variational Autoencoder model. Low dimensional data is used to train a Bayesian Multi Layer Perceptron. The proposed method outperformed the Bayesian Multi Layer Perceptron model which is trained on entire feature space in terms of accuracy performance. The accuracy of the proposed method is tested on baseline datasets.