Can Relational Learning Scale Up?
A key step of supervised learning is testing whether a can- didate hypothesis covers a given example. When learning in first order logic languages, the covering test is equivalent to a Constraint Satisfaction Problem (CSP). For critical values of some order parameters, CSPs present a phase pransition, that is, the probability of finding a solution abruptly drops from almost 1 to almost 0, and the complexity drama- tically increases. This paper analyzes the complexity and feasibility of learning in first order logic languages with respect to the phase transition of the covering test.
KeywordsPhase Transition Predictive Accuracy Information Gain Hard Problem Constraint Satisfaction Problem
Unable to display preview. Download preview PDF.
- [BGS99]M. Botta, A. Giordana, and L. Saitta. Relational learning: Hard problems and phase transitions. In Proceedings of th 16th International Joint Conference on Artificial Intelligence, pages 1198–1203, Stockholm, Sweden, 1999.Google Scholar
- [BGSS00]M. Botta, A. Giordana, L. Saitta, and M. Sebag. Relational learning: Hard problems and phase transition. In Selected papers from AIIA’99, volume to appear. Springer-Verlag, 2000.Google Scholar
- [GS00]A. Giordana and L. Saitta. Phase transitions in relational learning. Machine Learning, x:to appear, 2000.Google Scholar
- [HHW96]T. Hogg, B.A. Huberman, and C.P. Williams, editors. Artificial Intelligence: Special Issue on Frontiers in Problem Solving: Phase Transitions and Complexity, volume 81(1-2). Elsevier, 1996.Google Scholar
- [Qui90]R. Quinlan. Learning logical definitions from relations. Machine Learning, 5:239–266, 1990.Google Scholar