Home > Uncategorized > Structure-Learning By Dynamic Tree Bayesian Networks (DTBNs)

Structure-Learning By Dynamic Tree Bayesian Networks (DTBNs)

IN DTs, to search for a good structure is a very important issue. Changing the structure randomly will make the program so slow, we wneed to find some ways to come up with some good structures.

1) Map structure to structure space. This way when we search for structure, it is as if we are walking on the surface where each point on the surface is one unique structure.

2) After that we can perform any search algorithm on it, for example steepest decent, SA or even ITL. I think I would like to try to use QMI-ED to update the structure.

3) You can imagine the following. When you are standing on he surface, the height of the surface represents the log likelihood. YOu can only see the surface near to you, not too far away. Then from the information about the neighborhoods of the point you are standing at, you will make the next move based on the information for the neighborhood.

4) If we are at the point x on the surface, and somehow we can manage to get the neighborhood of x, ne(x). From x and ne(x), we can use an appropriate kernel to make a smooth pdf out of them. From the smooth surface, we can do whatever we want, for example we can do SA or we can use steepest decent, whatsoever method.

5) SA + kernel = kernel annealing is also an interesting option.

— some literatures

chemistry – molecular stability –> how to find a good structure of molecule
A new measure of edit distance between labeled trees (pdf)

read combinatorics, ask Adrian….structure space.

  1. No comments yet.
  1. No trackbacks yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: