Study of Pattern Storage Technique in Self Organizing Map using Hopfield Energy Function Analysis

Abstract

The Pattern Storage for the Continuum Features of the pattern can be characterized with the Self Organizing Map and Hopfield Energy Function Analysis. The Competitive Learning for the Self Organizing Map determines the Feature Mapping for the patterns with the Continuum Features. The iterations of the competitive learning between the Input Layer, the Feedback Layer reduce the neighboring region in the processing elements of Feedback Layer. On each iteration of this learning, the states of the feedback processing elements changes. The Energy Function corresponding to these states are determined. The change in Energy Function decreases; it shows that the network is approaches towards the Equilibrium State of the Global Stability. The minimum of the Energy States represents the stored pattern. The Network will able to encode the Pattern Information in the terms of Feature Space of the Patterns. Thus the pattern having the same feature will belong to the same Equilibrium State. This mechanism will help to determine the feature mapping for any unknown input pattern as well as any other prototype or noisy input pattern of the already stored pattern.