This paper investigates a new learning setting recently introduced by Vapnik (2006) that takes into account a known structure of the training data to improve generalization performance. This setting is a special case of a new inference technology known as learning with hidden information(Vapnik, 2006) suitable for many real-life applications with sparse high-dimensional data. We first briefly describe an extension of SVM called SVMgamma+ (Vapnik, 2006) that is associated with this new learning setting, and verify its effectiveness using a synthetic data set. Then we demonstrate the effectiveness of SVMgamma+ on a difficult real-life problem: detection of cognitive states from fMRI images obtained from different subjects. These empirical results show that the SVMgamma+ approach achieves improved inter-subject generalization vs standard SVM technology.
[1]
Vladimir Vapnik,et al.
Statistical learning theory
,
1998
.
[2]
Shai Ben-David,et al.
A theoretical framework for learning from a pool of disparate data sources
,
2002,
KDD.
[3]
Trevor Hastie,et al.
The Elements of Statistical Learning
,
2001
.
[4]
V. Vapnik.
Estimation of Dependences Based on Empirical Data
,
2006
.
[5]
Tong Zhang,et al.
A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data
,
2005,
J. Mach. Learn. Res..
[6]
D. Ruppert.
The Elements of Statistical Learning: Data Mining, Inference, and Prediction
,
2004
.
[7]
Tom M. Mitchell,et al.
Training fMRI Classifiers to Discriminate Cognitive States across Multiple Subjects
,
2003,
NIPS.