Sign in
Microsoft.com
 
Microsoft
 
 
 

locked
Forward-Backward Passes in Online Learning RRS feed

 > 

  • Question

  • Question
    Sign in to vote
    0
    Sign in to vote

    Hi,

    I'm working with a model that is very similar to the one at the end of http://social.microsoft.com/Forums/en-US/infer.net/thread/94bee5c7-056c-4933-81d0-cbc80bd4d888 to do online learning of TrueSkill for teams.  As noted in the TrueSkill Through Time paper (section 3.2), it would be nice to propagate information backwards in time, in addition to the forward pass that the other code performs.  I am having trouble figuring out how to do something like that with this model, since the framework (very usefully!) abstracts away the message passing discussed in the paper.  Can anyone provide insight or code that would get me started on this?

    Thanks!

    Friday, January 18, 2013 5:48 AM

All replies

  • Question
    Sign in to vote
    0
    Sign in to vote

    Propagating information forward and backward is actually easier in Infer.NET than propagating information only forward.  Instead of looping through games and running inference multiple times as in the linked code, you just create one big graph with all the games in it and run inference once. See How to represent large irregular graphs and TrueSkill Through Time.  By the way, in the TrueSkill Though Time paper, there was a concept of dynamics, where a player's skill could change over time.  This effect is not included in the linked code.

    Friday, February 1, 2013 10:35 AM
    Owner