1: \begin{abstract}

2: Acquiring labels are often costly, whereas unlabeled data are usually

3: easy to obtain in modern machine learning applications. Semi-supervised

4: learning provides a principled machine learning framework to address

5: such situations, and has been applied successfully in many real-word

6: applications and industries. Nonetheless, most of existing semi-supervised

7: learning methods encounter two serious limitations when applied to

8: modern and large-scale datasets: computational burden and memory usage

9: demand. To this end, we present in this paper the \emph{Graph-based

10: semi-supervised Kernel Machine} (GKM), a method that leverages the

11: generalization ability of kernel-based method with the geometrical

12: and distributive information formulated through a spectral graph induced

13: from data for semi-supervised learning purpose. Our proposed GKM can

14: be solved directly in the primal form using the Stochastic Gradient

15: Descent method with the ideal convergence rate $\text{O}\left(\frac{1}{T}\right)$.

16: Besides, our formulation is suitable for a wide spectrum of important

17: loss functions in the literature of machine learning (i.e., Hinge,

18: smooth Hinge, Logistic, L1, and $\varepsilon$-insensitive) and smoothness

19: functions (i.e., $l_{p}\left(t\right)=\left|t\right|^{p}$ with $p\geq1$).

20: We further show that the well-known Laplacian Support Vector Machine

21: is a special case of our formulation. We validate our proposed method

22: on several benchmark datasets to demonstrate that GKM is appropriate

23: for the large-scale datasets since it is optimal in memory usage and

24: yields superior classification accuracy whilst simultaneously achieving

25: a significant computation speed-up in comparison with the state-of-the-art

26: baselines. \keywords{Semi-supervised Learning \and  Kernel Method \and  Support Vector

27: Machine \and  Spectral Graph \and  Stochastic Gradient Descent} \vspace{-3mm}

28: \end{abstract}

29: