• Deep Learning
• Machine Learning
• Tree structured data
• Tree trasduction
• Variational Autoencoder

The thesis deals with the design of a deep learning model that can learn a generative process realizing unconstrained tree transductions. The model is based on an extension of the popular Variational Autoencoder framework to allow conditioning the generative process on tree structured inputs and to generate tree-structured predictions. It has been realized an efficient Tensorlow implementation of the proposed model, which has been validated on Arithmetic Expression trees and Neural Machine Translation