PyTorch Tutorial : Introduction to Neural Networks

Want to learn more? Take the full course at https://learn.datacamp.com/courses/introduction-to-deep-learning-with-pytorch at your own pace. More than a video, you'll learn hands-on coding & quickly apply skills to your daily work. --- In this lecture, we are going to introduce neural networks. More precisely, we are going to study fully connected neural networks, the simplest form of modern neural networks. In Supervised Learning with scikit-learn, you have seen some classifiers like k-nn. There are many other classifiers, some of which are very good, like Random Forests, Adaboost or Support Vector Machines. These classifiers work well when the data is given on vectorial format, as features. However, most of the data is not given as features. Instead, the data is in some rich format, like images, speech, text or video. In those cases, what people did before, was to use another algorithm to extract those features. In computer vision, the majority of last decade's research was on finding algorithms which get good features from images. Perhaps the most famous of those algorithms was the SIFT algorithm, which given an image, returns features from that image. Then those features are classified using a classifier like SVM. Maybe you can see the problem here. In order to solve the problem, we are optimizing two different algorithms (SIFT and SVM) which aren't related at all with each other. Neural networks work a bit differently. They have an input layer (in the figure denoted with 1), one or more hidden layers (denoted with 2), and an output layer (denoted with 3). The job of the hidden layers is to get good features, while the job of the output layer is to classify those features. With the net being trained end-to-end, we have a single algorithm which at the same time both finds good features, and classifies them. This has shown to work very well, and so neural networks have revolutionized many fields, to the point of making the old algorithms obsolete. Let's have