Brussels, Belgium Belgium
Dec. 10, 2012 to Dec. 10, 2012
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/ICDMW.2012.40
In practical engineering, small-scale data sets are usually sparse and contaminated by noise. It is difficult to guarantee a competitive generalization performance of regression model from such a data set. However, what is worth mentioning is that there are often a lot of incomplete relationships between attributes in practical engineering. The involvement of the relationships might be significant in improving the generalization performance of machine learning. So in this paper, we propose a transfer learning method based on the incomplete relationships between attributes, in which the incomplete relationships is reasoned to get complete relationships, and the complete relationships are then transferred to the regression learning to improve the generalization performance of the regression model. Finally the proposed method was applied to least squares support vector machine (LSSVM) and was evaluated on benchmark data sets. The experiment results show that the transfer learning can improve the generalization performance and prediction accuracy of the regression model.
Learning systems, Vectors, Machine learning, Robustness, Machine learning algorithms, Support vector machines, Benchmark testing, regression problem, relationships between attributes, Markov Logic Network, small sample, first-order predicate
Jinwei Zhao, Boqin Feng, Guirong Yan, Longlei Dong, "The Transfer Learning Based on Relationships between Attributes", ICDMW, 2012, 2013 IEEE 13th International Conference on Data Mining Workshops, 2013 IEEE 13th International Conference on Data Mining Workshops 2012, pp. 535-538, doi:10.1109/ICDMW.2012.40