Network Completion and Survey Sampling Steve Hanneke and Eric P. Xing Machine Learning Department Carnegie Mellon University.
Download
Report
Transcript Network Completion and Survey Sampling Steve Hanneke and Eric P. Xing Machine Learning Department Carnegie Mellon University.
Network Completion and Survey Sampling
Steve Hanneke and Eric P. Xing
Machine Learning Department
Carnegie Mellon University
I know the vertices, but
what does the network
look like?
The Task
Send out some
random surveys
Infer what the rest of
the network looks
like.
Survey:
“Who are you
linked to?”
Want a bound on number of pairs with incorrect predictions.
Outline
•
•
•
•
Notation
A Tight Bound
Relaxations of the bound
An Example
Notation
A Tight Bound
A Tight Bound
• The max in FT might be computationally hard
(open problem)
• Difficult to intuitively understand the behavior
• So we’d like to relax the bound
Relaxing the Bound
Relaxing the Bound
Relaxing the Bound
Proof
Example
Example
Conclusions
• In theory, can confidently learn a network
topology from survey samples
• Need a fairly strong learning bias
• Future: Extensions to Snowball Sampling?
Thanks