 |
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|
 |
 |
|
||||||||||
 |
|
|||||||||||
 |
|
|||||||||||
 |
 |
|
||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunchtime Chats
|
Fall 2005 Seminar
Date: December 2, 2005 Time: 10:00 AM - 11:00 AM Location: 3305 Newell-Simon Hall Speaker: Ajit Singh PhD Candidate Title: Finding Optimal Bayesian Networks by Dynamic Programming Abstract: Producing the Bayesian network structure that maximizes a score function is known as model selection. A common approach to model selection involves local search in the space of acyclic digraphs, which is prone to local maxima. Under a wide variety of assumptions the problem is NP-hard. Moreover, exhaustive enumeration requires super-exponential time in the number of variables. We present an exponential space/time algorithm for finding a structure that corresponds to the global maxima of a decomposable scoring function, such as BDeu or BIC. Joint work with Andrew Moore.