The first stage of ML is to train an ML model with examples. The form of machine learning that you'll be focused on in this specialization is called supervised learning, and in supervised learning we start from examples. An example consists of a label and an input. For example, suppose we want to train a machine learning model to look at images and identify what's in those images. The true answer is called the label. Cat for the first image and dog for the second image, those are the labels. The image itself, the pixels of the image are the input to the model. The model itself is a mathematical function of a form that can be applied to a wide variety of problems. There are many such mathematical functions. The models used in machine learning have a bunch of adjustable parameters though, all of them do. Then when we train a model, what we're doing is that we're making tiny adjustments to the model so that the output of the model, the output of the mathematical function, is as close as possible to the true answer for any given input. Of course, we don't do this on one image at a time. The idea is to adjust the mathematical function so that overall, the outputs of the model for the set of training inputs is as close as possible to the training labels. The key thing is that machine learning, at least machine learning of the form that we will consider in this course. The most mature from of machine learning relies on having a data set of labeled examples, and by labeled examples, we mean the input and the true answer. After the model is trained, we can use it to predict the label of images that it has never seen before. Here, we are inputting to the trained model, this image, and because the network has been trained, it is correctly able to output cat. Notice that the cat image on this slide is different from the one before it. It still works because the machine learning model has generalized from the specific examples of cat images that we showed it, to a more general idea of what a cat is and what it looks like. The key to making a machine learning model generalized data and lots and lots of it. Having labeled the data is a precondition for successful machine learning. It is important to realize that machine learning has two stages, training and inference. Sometimes people refer to prediction as inference because predictions seems to imply a future state. The case of images like this, we're not really predicting that it's a cat, just inferring that it's a cat based on the pixel data. It can be tempting as a data scientist to focus all your energy on the first stage on training. But this is not enough, you need to be able to operationalize the model, put the model into production so that you can run inferences. Look at many books on machine learning, blog posts, university courses, they tend to ignore the second stage of ML. But in the real world, what is the use of training a machine learning model if you cannot use it? In this specialization, we'll be careful to show you machine learning end-to-end, and by end-to-end, we mean putting machine learning models into production.
