G. Kunapuli, R. Maclin & J. Shavlik (2011).
Advice Refinement for Knowledge-Based Support Vector Machines. ICML 2011 Workshop on Combining Learning Strategies for Reducing Label Cost.
Poster (PDF).

This publication is contained in the following 2 PDF files
File 1, File 2.

Abstract:

Knowledge-based support vector machines (KBSVMs) incorporate advice from experts, which can improve accuracy and generalization significantly. A major limitation occurs when the expert advice is noisy or incorrect which can lead to poorer models and decreased generalization. We propose a model that extends KBSVMs and learns not only from data and advice, but also simultaneously improves the advice. This model, which contains bilinear constraints for advice refinement, is effective for learning in domains with small data sets. We propose two approaches to handle the bilinearity of the formulation along with experimental results.

Return to the publications of the Univ. of Wisconsin Machine Learning Research Group.