Tu sei qui:HomeNotizieEventiPoisson Sum-Product Networks: A Deep Architecture for Tractable Multivariate Poisson Distributions

Poisson Sum-Product Networks: A Deep Architecture for Tractable Multivariate Poisson Distributions

Seminari Prof. Kristian Kersting Dottorato di Ricerca in Informatica e Matematica
Quando 29/09/2016
dalle 15:30 alle 18:30
Dove Bari, Computer Science Department
Aggiungi l'evento al calendariovCal
iCal

Multivariate count data are pervasive in science in the form of histograms, contingency tables and others. Unfortunately,previous work on modeling this type of distributions do not allow for fast and tractable inference. In this seminar, I will present a novel Poisson graphical model, the first based on sum product networks, called PSPN, allowing for positive as well as negative dependencies. We present algorithms for learning tree PSPNs from data as well as for tractable inference via symbolic evaluation. With these, information-theoretic measures such as entropy, mutual information, and distances among count variables can be computed without resorting to approximations. Additionally, we show a connection between PSPNs and LDA, linking the structure of tree PSPNs to a hierarchy of topics. The experimental results on several synthetic and real world data-sets demonstrate that PSPN often outperform state-of-the-art while remaining tractable.

 

Based on joint works with Alejandro Molina, Sriraam Natarajan and Fabian Hadiji.

 

Azioni sul documento
Pubblicato il: 24/09/2016 
File allegato

Poisson.Kersting.pdf

Size 130.7 kB - File type application/pdf