Balance Support Vector Machines Locally Using the Structural Similarity Kernel

Abstract

Abstract. A structural similarity kernel is presented in this paper for SVM learning, especially for learning with imbalanced datasets. Kernels in SVM are usually pairwise, comparing the similarity of two examples only using their feature vectors. By building a neighborhood graph (kNN graph) using the training examples, we propose to utilize the similarity of linking structures of two nodes as an additional similarity measure. The structural similarity measure is proven to form a positive definite kernel and is shown to be equivalent to a regularization term that encourages balanced weights in all local neighborhoods. Analogous to the unsupervised HITS algorithm, the structural similarity kernel turns hub scores into signed authority scores, and is particularly effective in dealing with imbalanced learning problems. Experimental results on several benchmark datasets show that structural similarity can help the linear and the histogram intersection kernel to match or surpass the performance of the RBF kernel in SVM learning, and can significantly improve imbalanced learning results.

Object Templates for Visual Place Categorization

“…using background patches [21]. Furthermore, [7] illustrates that the structural similarity can greatly help to remove the ambiguities of pairwise similarity and provide more robust comparison results. Therefore, we use bibliographic coupling strength to measure the structural similarity of patches x 1 and x 2 , i.e., counting the number of patches that are nearest neighbors of both x 1 and x 2 .…”

“…We can further encode a local graph as a sparse neighborhood vector, and existing SVM training and testing techniques can be perform directly. [7] show that the structural similarity is more robust for examples with large intra-class variance and for imbalance dataset, making it suitable for our usage.…”

“…In Section 3.2, we present the global configurations approach by briefly reviewing [2] and describing a few improvements we make on it. In Section 3.3, we demonstrate how we utilize local information by selecting object templates with structure similarity SVM [7] from training images automatically and effectively, and classifying test images with these templates using Local NBNN [8].…”

“…[7] has shown that the structural similarity kernel is a valid SVM kernel. However, existing SVM algorithms cannot be applied directly applied to solve this kind of kernel SVM problem.…”

“…This method can automatically select important objects or key features of a semantic category based on training data by structural similarity SVM [7], and use them as templates for future Local Naive Bayes Nearest Neighbors [8] based classification on testing images. Combining this method with the global configurations approach, the accuracy is significantly improved.…”