Abstract

Unsupervised learning is very important in the processing of multimedia content as clustering or partitioning of data in the absence of class labels is often a requirement. In the recent work introduces a clustering algorithm based on two-phased neural network architecture. In the literature, neural network-based clustering methods mainly build a two-layer network with lateral inhibitive connections in the output layer. In the recent work how to find the optimum number of clusters and initial prototype vectors for a given unlabeled data set in the network architectures. To solve this problem proposed work, Bat Optimization Algorithm (BOA) is introduced in this work which is used to select optimum number of clusters and optimal selection of prototype vectors for a given unlabeled data set in the Unsupervised Hybrid Network Kernel Learning Support Vectors Clustering (UHNKLSVC). Here, new latent spaces are achieved in the hidden layers, and each layer can be used for a specific task in a greedy layer wise manner. Proposed work combine the procedures of an Autoencoder like network for unsupervised representation learning with the discriminative power of a Kernel Learning Support Vector Machine (KLSVM) network for fine-tuning the initial clusters by BOA. The initial network is referred as prototype encoding network, where the data reconstruction error is minimized in an unsupervised manner. Then clusters are selected via the use of BOA. The third phase, i.e., KLSVM network, endeavors to maximize the margin between cluster boundaries in a supervised way making use of the first output. Cluster fine-tuning is accomplished in a network structure by the alternate usage of the encoding part of both the networks. To conclude, the KLSVM network fine-tunes the previously established clusters in a unsupervised manner, and this BOA scheme enhances the clustering performance.

Paper Details