Optimized Weighted Samples Based Semi-supervised Learning

Through semi-supervised learning with graphs, the machine learning community has achieved many advantages in extracting information from a large volume of data under inadequate initial label information. Recent research has shown the benefit of weighing the samples that are labelled can give improved accuracy. Instead of providing similar consideration for labelled samples, sample weighting establishes higher weights for samples occupied at the border of multiple classes than labelled samples occupied so far from the boundary. This article proposes a faster way to calculate the sample weights by reducing the multiple clustering methods to single clustering. The new method of sample weighting is verified using a 2D feature set so that sample weighting can be easily visualized. The proposed method does not reduce the time complexity but it can reduce the number of steps required for weighting the samples. The obtained results have shown that this method can improve the speed with acceptable accuracy.