Desarrollo de una red neuronal con conexión hacia adelante para clasificar íconos.

 Informe de IA2

Realized for : Fabricio Gómez y Doménica Muñoz.

Proyecto Integrador Interciclo – Desarrollo de una red neuronal con conexión hacia adelante para clasificar íconos.

Basic concepts: 

The multilayer perceptron is an artificial neural network (ANN) made up of multiple layers, in such a way that it has the ability to solve problems that are not linearly separable, which is the main limitation of the perceptron (also called simple perceptron).

Multilayer Neural Network.

It is a generalization of the previous one, there is a set of intermediate layers between the input layer and the output layer (hidden layers). This type of network can be fully or partially connected(Figure 1).

Figure 1. Multilayer Neural Network.

2. The program should load the images from the given data corpus (which contains 25 categories of

animals: antelope, dog, kangaroo, tiger, etc.) considering that they are color images (RGB, Red,

Green and Blue) of  150 x 150 pixels (approximate), as can be seen in figure 2:

Figure 2. Images.

You will need to compare at least two different Convolutional Neural Network (CNN) configurations. For each neural network is necessary to carry out the following analysis: o Test at least 2 different neural network structures and activation functions, in order to determine which was the best alternative (considering number of neurons, number of layers and trigger functions).

The program must divide the corpus of images into two blocks: training (train) and tests

(test). Likewise, you must generate a report in PDF, Word or similar format that contains the following

information:

Precision plot (training data versus test data), 

Log loss graph (cost function: cross entropy), 

Accuracy and log loss comparison of the different neural network options you have tested.

Confusion matrix.

Developing

First we mount the disk in the Google Colab drive :

Figure 3. Disk mount on Google Colab.

 We load the data and import the libraries :

Fifure 4. Carga de datos.

We load the categories of the images :

Figure 5. Categories of the imges.

We manipulate the data by creating image arrays.