Multi Self-Organizing Map (SOM) Pipeline Architecture for Multi-View Clustering

Clustering has proved to be a successful classification method when it comes to dealing with multiview data. Each method and technique tries to achieve efficiency and accuracy in classifying the multiview data. Multi-source data contains noise and divergence. Another problem is that each view contains many features, so usually the multiview dataset is multi dimensional. This raises basic problems like the need for a dimensionality reduction technique for optimal selection of features, fusing the data of different views, and maintaining the inter- and intra-consensus of the multiview dataset. The fusion technique should merge the complementary information efficiently. The goal of this study is to use a promising technique for dimension reduction that reduces the noise but maintains the inter-view and cross-view consensus. A self-organizing map is one of the well-known unsupervised neural network algorithms used for preserving typologies during mapping from the input space (high-dimensional) to the display (low-dimensional).An algorithm called Local Adaptive Receptive Field Dimension Selective Self-Organizing Map 2 is a modified form of a self-organizing Map to cater different data types in the dataset. It calculates the dimension relevance with various data instances. These further place the relevant dimension samples in one group. The method does not need to know the number of clusters before hand, as it dynamically determines it during the process. Finally, this study proposed a novel multi-view learning framework that analyzes multi-source data and generates fine clusters efficiently.