1: \begin{abstract}

2: In this paper, we consider the harmonic extension problem, which is widely

3: used in many applications of machine learning. We find that the transitional

4: method of graph Laplacian fails to produce a good approximation of the classical

5: harmonic function. To tackle this problem, we propose a new method called the 

6: point integral method (PIM). We consider the harmonic extension problem from the

7: point of view of solving PDEs on manifolds. 

8: The basic idea of the PIM method is to approximate the harmonicity using 

9: an integral equation, which is easy to be discretized from points.

10: Based on the integral equation, we explain the reason why the transitional

11: graph Laplacian may fail to approximate the harmonicity in the classical sense

12: and propose a different approach which we call the volume constraint method (VCM). 

13: Theoretically, both the PIM and the VCM computes a harmonic function

14: with convergence guarantees, and practically, they are both simple, which amount

15: to solve a linear system.  One important application of the harmonic extension in 

16: machine learning is semi-supervised learning. We run a popular semi-supervised

17: learning algorithm by Zhu et al.~\cite{ZhuGL03} over a couple of well-known datasets 

18: and compare the performance of the aforementioned approaches. Our experiments show

19: the PIM performs the best. 

20: \end{abstract}

21: