ABSTRACT

Many of the machine learning tasks in computer vision may be viewed as a general regression problem where the goal is to figure out a mapping from some input data to some output data. This chapter conducts a thorough and formal exploration of regression as a mathematical tool. It presents linear regression models for this illustrative feature space. Further, the presents iterative approaches to finding a solution, which is useful for real problems where finding analytical solutions is too difficult or not possible at all. The chapter presents linear regression and the modeling of continuous valued labels. It uses the linearity assumption to create a linear model and parameterizes it. The chapter also expands upon the understanding of linear and nonlinear regression by solving the problem without an analytic solution. It uses the techniques of optimization, in particular the gradient descent, to solve the problem and find optimal weights.