Semi-Supervised Learning Based on Local Adaptive Kernels

Abstract

This paper presents a novel semi-supervised learning algorithm with local adaptive kernels named local feature adaptive learning (LFAL) for data classification. We employ multiple kernels to explore the local feature of data distribution and learn a novel Hilbert space from the kernel matrix of data derived from multiple kernels to extract more data information. Next, we construct a new semi-supervised learning problem with the learned Hilbert space under the general framework of semi-supervised learning. The advantage of the LFAL is that it learns more local data feature from data samples, and provides an adaptive way to learn the solution space with multiple kernels for learning problem. Experimental results on four real world data sets show the effectiveness of our algorithm.