1: \begin{abstract}

2: Owing to the prevalence of unlabeled data, semi-supervised learning

3: has been one of the most prominent machine learning paradigms, and

4: applied successfully in many real-world applications. However, most

5: of existing semi-supervised learning methods encounter two serious

6: limitations when applied to modern and massive datasets: computational

7: burden and memory usage demand. In this paper, we present the \emph{Graph-based

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

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

10: and distributive information formulated through a spectral graph induced

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

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

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

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

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

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

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

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

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

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

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

22: yields superior classification accuracy whilst simultaneously achieving

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

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

25: Machine \and  Spectral Graph \and  Stochastic Gradient Descent} 

26: \end{abstract}

27: