Hello, and welcome. In this video, we'll be reviewing a brief introduction to the convolutional neural network model. Convolutional neural networks or CNNs have gained a lot of attention in the machine learning community over the last few years. This is due to the wide range of applications where CNNs excel, such as object detection and speech recognition. In a later part of this video, we'll examine the object recognition problem, which is a key consideration when using CNNs. You'll also get an idea of how a CNN structure allows it to extract the elements from an image. For example, you'll see how a CNN learns to pick out a person or a dog or even chairs from an image, even if the chairs are only partially visible. Indeed, this is not an easy task for computers and it took years of dedicated research to achieve this. The original goal of the CNN was to form the best possible representation of our visual world in order to support recognition tasks. The CNN solution needed to have two key features to make it viable. First, It needed to be able to detect the objects in the image and place them into the appropriate category. Second, it also needed to be robust against differences in pose, scale, illumination, confirmation and clutter. The importance of this second feature simply cannot be overstated, since this was historically a huge limitation of hand-coded algorithms. Interestingly enough, the solution to the object recognition issue was inspired by examining the way our own visual cortex operates. In short, the CNN starts with an input image. It then extracts a few primitive features and combines those features to form certain parts of the object. Finally, it pulls together all of the various parts to form the object itself. In essence, it is a hierarchical way of seeing objects. That is, in the first layer, very simple features are detected. Then these are combined to shape more complicated features in the second layer and so on to detect things like a cat or a dog or any other object we're needing to detect. The convolutional neural network was developed from this sequence of steps. Let's consider the building in the lower right-hand corner of this image. How can a trained CNN learn and recognize that it's an image of a building, rather than some of the other elements that are situated in the photograph, such as a person or even those clouds in the sky? Well, in the training phase, CNNs would receive many building images as input. Then CNNs will automatically find that the best primitive features for a building would be things like horizontal and vertical lines. Much like the neurons firing in a person's visual cortex, the first layer in a CNN reacts when it receives an image that has horizontal or vertical lines. Once these simpler features such as lines are combined, CNNs learn to form higher abstract components, like the windows of the building and the building's external shape. It can then use these basic parts to form the complete object and learn how the entire building essentially looks like. Later, if it sees an image of a building that it hasn't seen before, CNNs can make a decision if the new input image is of a building based on the persistence of the various features it has stored. It is the same process for learning and detecting other objects, such as faces, animals, cars, and so on. So as you can see, the CNN is a set of layers with each of them being responsible to detect a set of feature sets, and these features are going to be more abstract as it goes further into examining the next layer. So at this point, you should have a basic understanding of the principles behind a convolutional neural network, and how CNNs might be used in an application. Thanks for watching.
