The paper presents a semi-supervised kernel method for regression estimation in the presence of unlabeled patterns. The method exploits a recently proposed data-dependent kernel which is constructed in order to represent the inner geometry of the data. This kernel is implemented into kernel regression methods (SVR, KRR). Experimental results aim to highlight the properties of the method and its advantages as compared to fully supervised approaches. The influence of the parameters on the model properties was evaluated experimentally. One artificial and two real-world datasets were used to demonstrate the performance of the proposed algorithm
[1]
Thorsten Joachims,et al.
Transductive Learning via Spectral Graph Partitioning
,
2003,
ICML.
[2]
Vladimir Vapnik,et al.
Statistical learning theory
,
1998
.
[3]
Mikhail Belkin,et al.
Problems of learning on manifolds
,
2003
.
[4]
Alexander J. Smola,et al.
Learning with kernels
,
1998
.
[5]
Alexander Zien,et al.
Semi-Supervised Classification by Low Density Separation
,
2005,
AISTATS.
[6]
Mikhail Belkin,et al.
Beyond the point cloud: from transductive to semi-supervised learning
,
2005,
ICML.