A Third Order based Additional Regularization in Intrinsic Space of the Manifold

Rakesh Kumar Yadav (Indian Institute of Information Technology-Allahbad)
Abhishek Singh (Indian Institute of Information Technology-Allahbad)
Shekhar Verma (Indian Institute of Information Technology-Allahbad)
S. Venkatesan (Indian Institute of Information Technology-Allahbad)
M. Syafrullah (Universitas Budi Luhur)
Krisna Adiyarta (Universitas Budi Luhur)

Proceeding of the Electrical Engineering Computer Science and Informatics
Vol 6: EECSI 2019

Second order graph Laplacian regularization has the limitation that the solution remains biased towards a constant which restricts its extrapolationcapability. The lack of extrapolation results in poor generalization. An additional penalty factor is needed on the function to avoid its over-fitting on seen unlabeled training instances. The third order derivative based technique identifies the sharp variations in the function and accurately penalizes them to avoid overfitting. The resultant function leads to a more accurate and generic model that exploits the twist and curvature variations on the manifold. Extensive experiments on synthetic and real-world data set clearly shows thatthe additional regularization increases accuracy and generic nature of model.