Computer Vision News - December 2020

Computer Vision Tool 10 representation of the input which is called a map. The main difference to the often used ANNs is that instead of back-propagation they use graphs to indicate the properties of the input space. To explain it, we’ll take a simple dataset, the famous Iris from R.A. Fisher. You can easily find and download this dataset, but it is also available through the Scikit- Learn Python package. In that case it can be loaded as follows: Self-Organizing Maps (SOMs) with PyTorch Welcome back! :) A term which isn’t usually heard in the field of Neural Networks is that of a self-organizing map (SOM). Although it’s difficult to describe it briefly, a SOM is a type of Artificial Neural Network (ANN) which is used as way to reduce dimensionality. This is achieved by creating a discretized by Ioannis Valasakis (@wizofe) The Python program that follows creates a 40x40 SOM of the dataset as a data structure. In the Figure 1, a U-Matrix is created with the black cells showing the similarity of the items and the white cells showing the borders between the item clusters. It can be shown that 3 different clusters (classes) are defined. T he dataset has four dimensions and there are three different labels (numbered as 0, 1 and 2). In order to visualize the dimensions, the SOM is used to reduce the dimensions to two. Creating SOMs using Python The Iris dataset includes 150 items. Some representative lines are shown below: from sklearn import datasets # load R.A. Fisher’s Iris dataset iris = datasets.load_iris() 5.1,3.5,1.4,0.2,0 4.9,3.0,1.4,0.2,0 7.0,3.2,4.7,1.4,1 6.3,3.3,6.0,2.5,2