Introduction to Convolutional Neural Networks

Table of Contents

1. Introduction
2. Handwritten Digit Recognition
3. Artificial Neural Networks
4. Deep Learning Series
5. Image Classification
6. Convolutional Neural Networks
7. Feature Extraction
8. Value and Pooling Operations
9. Benefits of Convolutional Neural Networks
10. Handling Rotation and Thickness
11. Automatic Filter Detection
12. Conclusion

Convolutional Neural Networks: A Beginner's Guide

Convolutional Neural Networks (CNNs) are a Type of artificial neural network that are commonly used for image classification tasks. In this article, we will provide a simple explanation of CNNs without using much mathematics, so that even a high school student can understand it easily.

Introduction

Let's say You want the computer to recognize a handwritten digit, such as the number 9. The way a computer looks at this is as a GRID of numbers. In reality, it will use RGB numbers from 0 to 255. The issue with this presentation is that it is too hard-coded. If there is a little shift in the digit 9, for example, if the 9 is in the middle in one case and on the left in another case, the representation of numbers just changes and it doesn't match with our original number grid. As a result, the computer will not be able to recognize that this is the number nine. There could be a variation in how you write the digit, which will change the two-dimensional representation of numbers, and again you will not be able to match it with the original grid. This is where artificial neural networks come in.

Handwritten Digit Recognition

In a previous video on artificial neural networks for handwritten digit recognition, we created a one-dimensional array by flattening the two-dimensional representation of our handwritten digit number. Then we built a neural network with one Hidden layer and output layer. This dense neural network will work okay for a simple image like a handwritten digit, but when you have a bigger image, such as a picture of a Koala, it becomes too much computation for your computer.

Artificial Neural Networks

The disadvantages of using an artificial neural network for image classification are too much computation and treating local pixels the same as pixels far apart. If you have a koala's face in the left corner versus the right corner, it is still a koala, and it doesn't matter where the face is located. However, image recognition tasks are centered around locality. Okay, so if the pixels are moved around, it should still be able to detect the object in an image, but with an artificial neural network, it's hard.

Deep Learning Series

In this deep learning series, we have already looked at artificial neural network video on handwritten digit recognition. If you have not seen that video, please make sure you see it so that your fundamentals on artificial neural networks are clear.

Image Classification

The main idea behind convolutional neural networks is feature extraction because the Second part is the same, which is a simple artificial neural network. By doing this convolution, you are detecting the features, and you are also reducing the dimension. There are three benefits of the convolution operation. The first one is connection sparsity, which reduces overfitting. Connection sparsity means not every node is connected with every other node like in an artificial neural network, where we call that a dense network. Here we have a filter which we move around the image, and at a time, We Are talking about only a local region, so we are not affecting the whole image.

Convolutional Neural Networks

The second benefit is convolution and pooling operation combined gives you a location invariant feature detection, which means koala's eye could be in the left corner or the right corner, anywhere, and it will still detect it. The third is parameter sharing, which is when you learn the parameters for a filter, you can Apply them in the entire image. The benefit of value is that it introduces non-linearity, which is essential because when we are solving deep learning problems, they are non-linear by nature. It also speeds up training and is faster to compute.

Feature Extraction

Filters are nothing but feature detectors. For koala's case, you can have an eye detector, and when you apply convolution operation, in the result, you see that you got these two eyes at this location. If the eyes are at a different location, it will still detect because you are moving the filter throughout the image, and they are location invariant, which means it doesn't matter where the eyes are in the image; these filters will detect those eyes and activate those particular regions.

Value and Pooling Operations

The benefit of pooling is that it reduces dimension and computation, reduces overfitting, and makes the model tolerant to small distortions. There are two types of pooling operations: max pooling and average pooling. Max pooling is more generally used, but sometimes people use average pooling also.

Benefits of Convolutional Neural Networks

The benefits of convolutional neural networks are connection sparsity, location invariant feature detection, parameter sharing, non-linearity, and reduced computation.

Handling Rotation and Thickness

By itself, CNN cannot handle rotation and thickness, so you need to have training samples that have some rotated and scaled samples. You can use data augmentation techniques to generate new samples.

Automatic Filter Detection

The most fascinating part about neural networks, in general, is that they will automatically detect these filters on their own. That is part of the training. When the neural network is training or when the CNN is training, because you're supplying thousands of koalas' images here, using that, it will use backpropagation, and it will figure out the right amount of filters. It will figure out the values in this filter, and that is part of the learning or the backpropagation.

Conclusion

In conclusion, convolutional neural networks are a type of artificial neural network that are commonly used for image classification tasks. They are beneficial because they reduce computation, reduce overfitting, and make the model tolerant to small distortions. They are also capable of detecting features and handling rotation and thickness. The most fascinating part about CNNs is that they will automatically detect these filters on their own.