Classifier with hierarchical topographical maps as internal representation

Abstract

In this study we want to connect our previously proposed context-relevant topographical maps with the deep learning community. Our architecture is a classifier with hidden layers that are hierarchical two-dimensional topographical maps. These maps differ from the conventional self-organizing maps in that their organizations are influenced by the context of the data labels in a top-down manner. In this way bottom-up and top-down learning are combined in a biologically relevant representational learning setting. Compared to our previous work, we are here specifically elaborating the model in a more challenging setting compared to our previous experiments and to advance more hidden representation layers to bring our discussions into the context of deep representational learning.

Rate Increase of the Objects Classification on the Convolutional Neural Networks With the Self-Organization Maps Implementation

“…Traditionally, the additional circuitry in combination with the SOM uses a multi-layered perceptron. In modern studies [10][11][12], SOM has been connected to convolutional networks. These studies aim to improve SOM clustering.…”

“…In [11], the concept is to train multiple SOM, each corresponding to a separate area of the input image. In [12], the hidden layers are replaced by modified self-organizing maps. However, all these studies are aimed at improving the efficiency of Kohonen's self-organizing maps.…”