Abstract

2. Setting
In the joint inference setting, we seek to estimate the distribution P (Y| X) and determine the most likely values of target variables Y based on observations X. We represent this probability distribution with a Markov random field (MRF), and express variable interactions with a set of logical rule templates. R refers to groundings of these rules generated by atoms in Y and X. The MRF can be decomposed into a set of weighted potentials corresponding to these ground rules, φr (I) having weight wr, where I is an assignment of the variables Y: f (I)= 1

Date

2013

Authors

Jay Pujara, Hui Miao, Lise Getoor

Journal

ICML Workshop on Machine Learning with Test-Time Budgets