CNN MNIST

CNN MNIST

Intro

This model is improved version of simple MNIST model (with just 1 layer).
For example, model will classify above image as 2.
Simple MNIST model's accuracy was about 92% but this model can increase up to 99.2%.
Then how this improved MNIST model can increase accuracy?

What is CNN?

This improved model use CNN to extract feature. When input data of ANN model is image, it's hard to select feature of image manually. To improve accuracy of ANN model when input data type is image, we can use ANN to extract feature of image automatically. CNN is basic technology to extract feature in image. In MNIST problem, input data is image of handwritten number, So we can improve accuracy by using CNN to extract feature

CNN is consist of convolution layer and pooling layer. Convolution layer extract feature of image by matrix multiplication. Pooling layer is used to reduce feature size or emphasize specific feature.

CNN can improve accuracy of tensorflow model which handle image data. The reason of improvement is, Fully connected layer only support 1D data and image is 2D data. We need to flatten 2D image to 1D data to use fully connected layer. In this step, spatial information of 2D image is lost. To avoid lost of spatial informations, we use CNN which extract spatial information as feature. So when we use spatial data as input, we can improve accuracy by using CNN

Architecture

Above picture depict architecture of CNN MNIST model. First layer is input layer which accept user input. Input layer accept user input as [?, 28 * 28] list. Input layer forward this value to CNN layer.

From 2nd layer to 5th layer is CNN layer. One CNN layer is consist of 1 convolution layer and 1 pulling layer. So we can say that there are 2 CNN layers. Second CNN layer forward extracted feature to ANN.

6th and 7th layer is ANN layer. This layer using feature which is extracted from CNN layer and classify number. 6th layer using ReLu as loss function and 7th layer using Softmax as loss function.

Last layer is called output layer, There are 10 neurons and the index of neuron which has the largest value will be the prediction of number from 0 to 9.

Upload

Upload Image! ML will detect your handwritten number.

Please upload image which contains 1 number, This AI is stupid...

About

Lorem ipsum dolor sit amet, consectetur et adipiscing elit. Praesent eleifend dignissim arcu, at eleifend sapien imperdiet ac. Aliquam erat volutpat. Praesent urna nisi, fringila lorem et vehicula lacinia quam. Integer sollicitudin mauris nec lorem luctus ultrices. Aliquam libero et malesuada fames ac ante ipsum primis in faucibus. Cras viverra ligula sit amet ex mollis mattis lorem ipsum dolor sit amet.

Contact

• Twitter
• Facebook
• Instagram
• GitHub

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';