Politecnico di Torino (logo)

Variational Autoencoder for unsupervised anomaly detection

Francesco Lupo

Variational Autoencoder for unsupervised anomaly detection.

Rel. Paolo Garza. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Matematica, 2019

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (4MB) | Preview

The variational autoencoder is a generative model that is able to produce examples that are similar to the ones in the training set, yet that were not present in the original dataset. While this model has many use cases in this thesis the focus is on anomaly detection and how to use the variational autoencoder for that purpose. In the first part various state of the art anomaly detection algorithms are presented, in the second part the structure and the functioning of the variational autoencoder are presented, along with a comparison with the classic autoencoder. The details on how to exploit the variational autoncoder as an anomaly detection tool are also described in this part. In the third part experiments carried out with different datasets and different architectures are shown. In the last part a new use case is proposed, in particular on how to use the variational autoencoder to perform semantic novelty detection in a natural language processing context.

Relators: Paolo Garza
Academic year: 2018/19
Publication type: Electronic
Number of Pages: 67
Corso di laurea: Corso di laurea magistrale in Ingegneria Matematica
Classe di laurea: New organization > Master science > LM-44 - MATHEMATICAL MODELLING FOR ENGINEERING
Aziende collaboratrici: Reply Consulting Srl
URI: http://webthesis.biblio.polito.it/id/eprint/10360
Modify record (reserved for operators) Modify record (reserved for operators)