SRL2004: Statistical Relational Learning and
Statistical machine learning is in the midst of
revolution". After many decades of focusing on independent and
identically-distributed (iid) examples, many researchers are now
studying problems in which the examples are linked together into
complex networks. These networks can be a simple as sequences and 2-D
meshes (such as those arising in part-of-speech tagging and remote
sensing) or as complex as citation graphs, the world wide web, and
relational data bases.
Statistical relational learning raises many new challenges and opportunities. Because the statistical model depends on the domain's relational structure, parameters in the model are often tied. This has advantages for making parameter estimation feasible, but complicates the model search. Because the "features" involve relationships among multiple objects, there is often a need to intelligently construct aggregates and other relational features. Problems that arise from linkage and autocorrelation among objects must be taken into account. Because instances are linked together, classification typically involves complex inference to arrive at "collective classification" in which the labels predicted for the test instances are determined jointly rather than individually. Unlike iid problems, where the result of learning is a single classifier, relational learning often involves instances that are heterogeneous, where the result of learning is a set of multiple components (classifiers, probability distributions, etc.) that predict labels of objects and relationships between objects.
There have been several workshops on relational learning in recent years. The goal of this workshop is to reach out to related fields that have not participated in previous workshops. Specifically, we seek to invite researchers in computer vision, spatial statistics, social network analysis, language modeling and probabilistic inference to attend the workshop and give tutorials on the relational learning problems and techniques developed in their fields.
Because our goal is to build links with other fields, a significant amount of time in the workshop will be devoted to invited tutorials and discussion. Tutorials will include (a) overview of relational learning, (b) relational learning in spatial statistics, (c) relational learning in social network analysis, (d) relational learning in computer vision, and (e) approximate probabilistic inference for large networks. Focus topics for discussion will include (a) methodology (e.g., how to evaluate machine learning research on linked data by using connections between the test set and training set in a principled manner). (b) barriers to progress (the cost of inference, the need for benchmark data sets), and (c) new application directions (short talks describing interesting new applications). To give participants an opportunity to share their research with others, we plan to have at least one poster session
Participation is open. Participants are encouraged to submit papers (maximum 6 pages). Accepted papers may be presented orally or as posters.
April 2 Workshop papers due
Tom Dietterich, Oregon State University
James Cussens, University of York, UK