Markov Logic Networks with Grounding-Specific Weights
A limit of Markov logic is the difficulty of handling continuous features: this problem can be crucial when dealing with tasks in which the use of such continuous features is necessary (e.g. protein profiles in bioinformatics). We extended Markov logic networks by letting different groundings of a same first-order formula to have different weights. The weight of each ground formula can be computed as a function of its constants: our implementation uses neural networks to compute these weights. This approach also allows us to incorporate continuous features in a natural way.
We implemented Markov logic networks with grounding-specific weights in a modified version of the Alchemy system.
Here you can download a ZIP file containing the binaries of our software, and some example files.
A small manual is also available here for download.
For any questions or comments, please contact:
lippi AT dsi DOT unifi DOT it
Back to Marco Lippi's web page