Clustering with variational autoencoders

Details

Supervisors

Hainaut, Donatien

Faculty

Faculté des sciences

Degree label

Master [120] en science des données, orientation statistique, à finalité spécialisée

Abstract

Autoencoders are neural networks used for dimensionality reduction and for embedding categorical variables. The analysis of encoded signals with clustering (K-means) or visualization (T-SNE) algorithms allows to detect hidden structures in datasets. In this thesis, we will focus on variational autoencoders. They learn a distribution over the latent data instead of a discrete latent space. This presents several advantages. Compared to classical autoencoders, the compressed signals are less correlated. Secondly, they allow to generate fake datasets. In this thesis, we will first implement variational autoencoders in Python-Keras and next apply unsupervised learning algorithms to detect clusters. Finally, we will try to generate deep fake datasets.