Gastvortrag von Davide Bacchiu

Public Event
01. August 2018

Learning generative models for structured data

Abstract – Structured data allows representing complex compound information made of atomic information pieces, the vertices and their labels, enriched by complex relational information, represented by the edges. Recently there has been a renewed interest on adaptive processing of structured data, thanks to the roaring attention of the deep learning community. In this talk, I will discuss how generative models can be used to process structured data and to model complex distribution over structured samples. I will start by introducing the recursive extension of hidden Markov models for directed acyclic data, showing how these can be used to define efficient adaptive tree kernels. Then, we will see how the model can be extended to deal with general classes of graphs through a contextual approach, introducing a deep generative model for graphs.