Data:  08/10/2015
 Horário: 14hs
  Local: Auditório do CCSL

 Palestrante: Walter Perez
  Título: Initialization Heuristics for Greedy Bayesian Network Structure Learning

 Resumo:
 Bayesian networks have become one of the most popular methods in the modeling of uncertain knowledge. Manually specifying a Bayesian network over a large and complex domain is a labour-intensive, error-prone task. Thus, techniques that learn Bayesian networks exclusively from data have become indispensable. A popular and effective approach for learning Bayesian network structures is to perform a greedy search on the space of variable orderings followed by an exhaustive search over the restricted space of compatible parent sets. This so-called Order-Based Greedy Search is usually initialized with a randomly sampled order. While this guarantees an unbiased coverage of the space of ordering, it can lead to poor local optimal and slow convergence issues. In this talk, I will present some techniques for generating informed solutions to Order-Based Greedy Search of Bayesian network structures. I will briefly review the basics of Bayesian networks, their applications, the learning structure problem and the main approaches to solve it. Then I will discuss the novel techniques for generating initial solutions we devised, and show some experimental results obtained on real-word data.

Todos são benvindos!