[SOUND] Hello, I'm standing next to the Upwells sculpture on the north quad at the University of Illinois. This sculpture consists primarily of a sundial and can be used to monitor the changing of the seasons, estimate the time of day, calculate the latitude and make other astronomical measurements. Upwells has one large gnomon that forms the spire of the sundial. There are also three domes on which the shadow of the gnomon falls in special times of the year. These are the summer solstice, the spring and autumnal equinoxes and the winter solstice in order of the longest shadow. The gnomon can also be used at night to identify Polaris, the North Star. The earliest known sundials are believed to be around 3,500 years old, first appearing in ancient Egypt. Sundials thus represent one of the earliest attempts by humans to model the natural world. In this course, we are performing the modern equivalent. We are using machines to generate models that learn from data. These models can either be used to make predictions for new unseen data. Or they can be used to provide a deeper understanding of some process. The first module for this course provides a foundation for machine learning on which you will build in the rest of this course. First, you will be introduced to how machine learning is used in business and accountancy. The modern world of accounting is changing rapidly, data is everywhere. And efficiently using machines to help make sense of this growing wealth of data is becoming increasingly important. This course will prepare you to learn from data effectively today, and also to prepare you for an uncertain future where artificial intelligence will become even more integrated into our daily lives. Next, you will learn about the basic types of machine learning. This starts with a discussion on the basic ways that machines can generate models from data. The first method is supervised learning, where labeled training data are used to guide an algorithm to generate a model. The second method is unsupervised learning, where an algorithm learns patterns directly from the data. In some cases, a combination of these two methods is employed, where the algorithm learns patterns that are then quantified by using training data to understand the patterns. In this course, we will use the Python programming language to perform data analytic tasks. This includes applying machine learning to data for which we will use the scikit-learn library, one of the most popular machine learning libraries in existence. You will learn in this module how to effectively use the four main types of machine learning. Classification, regression, clustering and dimensional reduction. Briefly, for classification we build a model to predict one of a set of discrete classes. While for regression, we predicted a continuous value. Clustering involves identifying structure in a data set and using this structure to predict which group an instance belongs to. While dimensional reduction aims to generate new features from the original data that can simplify the task of learning more complex models. To actually use one of these algorithms, special parameters, known as hyperparameters, must be specified. These hyperparameters cannot be learned from the data directly. And are instead determined by identifying the set of hyperparameter values that produce the optimal results for our machine learning task. Machine Learning can seem daunting when first encountered. Thus, to demonstrate how to apply machine learning to a specific problem, you will next learn how to treat linear regression as a machine learning problem. This will include a discussion of cost functions, which underly machine learning models which seek to minimize that cost function in order to find the optimal model for a given data set. Linear models, while simple, are very powerful, since they are easy to understand and to explain to others. Finally, you will learn about one of the simplest machine learning algorithms, the k-nearest neighbor algorithm. This algorithm leverages the fact that data that are similar are often co-located. You can see this in the world around you. Most people live near people who are similar to them. Either financially, ethnically, or in political or religious views or all of the above. We can leverage existing relationships in the data to average over data in a neighborhood to predict values for new instances. As you will see, this algorithm can be used for both classification and regression. Now, our course server is set up to support you in applying these algorithms on different data sets. As this material may likely be new, be sure to stay on top of the content and ask questions as they arise. I hope you are excited to get started. This is the first step in learning to build models from data, good luck.
