I have gone through the wikipedia explanation of SRL. But, it only confused me more:
Statistical relational learning (SRL) is a subdiscipline of artificial intelligence and machine learning that is concerned with domain models that exhibit both uncertainty (which can be dealt with using statistical methods) and complex, relational structure.
Can someone give a more dumbed down explanation of the same, preferably with an example?
The University of Maryland published some slides (PDF) from an introductory presentation on this topic.
The fourth page explains why SRL is interesting. "Traditional statistical machine learning approaches" process one sort of thing in which there is some uncertaintly. Image identification is a good example of that. "Traditional ILP/relational learning approaches" use several kinds of information to produce hypotheses about the data set, but apparently allow for no noise in the data.
Statistical relational learning models are intended to work with data sets that have several types of objects connected to each other via various links (hence "relational"). They also are meant to deal with uncertainty (hence "statistical").
Skipping past some slides that aren't really useful without a transcript of what was said over them, we come to slide 17, which has comprehensible definitions and examples:
- Object classification
- Predicting the category of an object based on its attributes and its links and attributes of linked objects
- e.g., predicting the topic of a paper based on the words used in the paper the topics of papers it cites the used in the paper, the topics of papers it cites, the research interests of the author
- Object type prediction
- Predicting the type of an object based on its attributes and its links and attributes of linked objects
- e.g., predict the venue type of a publication (conference, journal, workshop) based on properties of the paper
As you can see, these models can keep track of several things and the interactions between them. The next slide talks about link prediction, the ability to predict several attributes of connections between objects, like the importance/quality of a citation. As previously mentioned, these models don't require 100% accurate data to give interesting results; academic citation lists might occasionally be less than comprehensive, and the importance of a citation is challenging to quantify.
Like ILP, SLP will hopefully be able to "see" new kinds of links between "entities", as with the presentation's example of identifying research communities.
Past slide 20, the presentation goes into some serious mathematics. It does have a much less technical conclusion starting at slide 198.
