Abstract:A similarity or dissimilarity measure, such as the Euclidean distance, is crucial to discriminative clustering algorithms. These measures used to calculate pairwise similarities between samples rely on data representations in a feature space. However, discriminative clustering fails if the samples in a feature space are linearly inseparable. This problem can be solved by performing a nonlinear data transformation into a high dimensional feature space, which can increase the probability of the linear separability of the samples within the transformed feature space and simplify the associated data structure. Mercer kernels, which are constructed using such a nonlinear data transformation, have been widely used in clustering tasks. Extreme learning machine (ELM) is a new method that exhibits promising clustering performance owing to its universal approximation capability, easy parameter selection, explicit feature mapping process, and excellent feature representation capability. This study proposes an ELM based multi-view learning approach with different views generated by ELM random feature mapping with respect to different hidden-layer nodes, and exploits the properties of these views. Experiments show that better clustering results can be obtained by combining these views together compared with the corresponding ELM-based single-view clustering methods and the traditional algorithms which are performed in the feature space of the original data. Moreover, local kernel alignment property is widespread in these views. This alignment helps the clustering algorithm focus on closer sample pairs. This study also proposes an ELM based multiple kernel clustering algorithm with local kernel alignment maximization. The proposed algorithm is experimentally demonstrated on 10 single-view benchmark datasets and yields superior clustering performance when compared with the state-of-the-art multi-view clustering methods in recent literatures. Thus, the effectiveness and superiority of maximizing local kernel alignment on those views constructed by the proposed method are verified.

Local extreme learning machine: local classification model for shape feature extraction

Correntropy Extreme Learning Machine Based on Spatial Pyramid Matching and Local Receptive Field

Local Deep Feature Learning Framework for 3D Shape.

Local Receptive Fields Based Extreme Learning Machine with Hybrid Filter Kernels for Image Classification

Image Classification Using Low-Rank Regularized Extreme Learning Machine

3d Shape Segmentation And Labeling Via Extreme Learning Machine

Extreme Learning Machine with Multi-Scale Local Receptive Fields for Texture Classification

Quasi-curvature Local Linear Projection and Extreme Learning Machine for Nonlinear Dimensionality Reduction

Discriminative Globality-Locality Preserving Extreme Learning Machine for Image Classification.

Extreme Learning Machine with Kernel Model Based on Deep Learning

Local Kernel Alignment Based Multi-View Clustering Using Extreme Learning Machine

Orthogonal Extreme Learning Machine for Image Classification

Enhanced LLE Features Classification for Face Recognition

A Novel Image Classification Algorithm Based on Extreme Learning Machine

Face Recognition Based on LBP and Extreme Learning Machine

Discriminative Locally Linear Embedding in Image Space

Face recognition via local preserving average neighborhood margin maximization and extreme learning machine

Supervised Classification of Hyperspectral Images Using Local-Receptive-fields-based Kernel Extreme Learning Machine.

Clustering in extreme learning machine feature space

Local Linear Transformation Embedding

An ELM Based Local Topology Preserving Hashing