An Artificial Neural Network Model for Multi Dimension Reduction and Data Structure Exploration

Abstract

This paper proposes an hybrid Artificial Neural Network (ANN) with Self-Organizing Map (SOM) and modified Adaptive Coordinates (AC) for multivariate dimension reduction and data structures exploration. SOM, being a prominent unsupervised learning algorithm, is often used for multivariate data visualization. However, SOM only preserved input space inter-neurons distances and not in the output space because of SOM rigid grid. SOM grid provides little information for visual exploration of the clustering tendency of the multivariate data. Modified AC is therefore proposed to remove SOM's map rigidity and provides better data topology preserved visualization. Empirical study of the hybrid yielded promising topology preserved visualizations for synthetic and benchmarking datasets.

A Hybrid Visualization-Induced Self-Organizing Map for Multi Dimensional Reduction and Data Visualization

“…As highlighted in [8] and [9], AC suffers from inconsistent adaptive units movements due to the use of relative adaptation factor as shown in Eq. (7).…”

“…(7). In order to successfully hybridize modified AC into SOM while retaining SOM robustness, an extra set of coordinates ax i ,ay i  are used as the adaptive units [9]. In order to successfully hybridize modified AC into SOM while retaining SOM robustness, an extra set of coordinates ax i ,ay i  are used as the adaptive units [9].…”

“…AC is similar to SOM except it uses virtual adaptive units to mirror SOM neurons movement. Modified AC as reported in [9] is therefore proposed to be hybridized with the ViSOM algorithm in order to alleviate this drawback. On the other hand, ViSOM's topological and inter-neuron distances preserved map is produced at the cost of larger number of neurons although most of these neurons are never activated.…”

A hybrid Spiking Neural Network model for multivariate data classification and visualization

“…Solving spatial problems in temporal domain introduces unnecessary complications. Therefore, this study proposes a hybrid learning model based on Self Organizing Map with modified Adaptive Coordinates (SOM-AC) [19] and SNN. SOM is the most popular unsupervised ANN learning algorithm in the market.…”

“…However, original SOM based ANN or SNN architecture is unable to provide the system user with such information. Therefore, modified adaptive coordinate [19] is hybridized into the learning algorithm to produce an intuitive, topologically preserved visualization. This proposed hybrid model can accommodate for both spatial and temporal data with simple transformation from one form to the other.…”