To Begin

Deep learning is one of the domains of machine learning, based on “Neural Networks”.

What is “Neural Network”?

Literally, neural network means a bulk of neurons (or called nodes) connect to each others and string to a network. Each neuron transmits information to the next neuron one by one, and the connections between them are computational relationships as the figure shown.

A result will be generated once something is input the network. Lenet, which is deem as the success of using convolution neural networks in computer vision, predicts the handwritten digits by feeding images to the network.

As networks become larger, we call it “deep”.

Why is “convolution”?

Convolution is a mathematic operation by weight and sum. Convolution in the two-dimensional space amplifies the features in the original image and also elimiates the noises. Therefore, we are able to use the extracted features to recognize images. We can see the edges of items in the image appear when applying convoluton. There is the demo website.

What is “learning”?

As we know, there are mathematic relations between all the neurons. Our purpose is to figure out the best parameter combination in networks which manage to produce a nice result. To do so, we put the ground truth and the result produced from the network to formula a “cost function”. As the cost converges, the ideal reult is closer. We call the process “learning”.

Reference

•Figure1: http://tech.huanqiu.com/discovery/2013-01/3572680.html

•Figure2: http://www.bogotobogo.com/python/scikit-learn/Artificial-Neural-Network-ANN-1-Introduction.php

•Figure3: http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf

If you like this article and consider it useful for you, please support it with 👏.