MINST-NN-and-CNN-models

Creating models for MINST database

The MINST database is formed by digits images with labels.

Problem

Build and traing a classifing model for correct classify the digits from 1 to 10. The images are made by 28x28 pixels in one channel color gray color.

Solutions

I provided two different model to solve this classification task. The solutions are:

1. Full connected NeuralNetwork
2. Convolutional Neural Network

Full connected NeuralNetwork

The Neural Network id made by:

• 784 nodes in input
• 50 nodes in the Hidden Layer
• 10 output nodes

784 nodes are caused by the dimension of the image 28x28 = 784, we reshape a matrix(28,28) to an array(784).

• Training: Got 55786 / 60000 with accuracy 92.98%
• Test: Got 9295 / 10000 with accuracy 92.95%

Convolutional NeuralNetwork

- Training: Got 59089 / 60000 with accuracy 98.48% - Test: Got 9842 / 10000 with accuracy 98.42%