Machine Learning is, to a large extent, the process of acquiring abstractions of the real world from a sparse set of observations. The observations can include software, webpages, DNA and protein arrays, motion capture data, images, computer game logs, music, video, controlled simulations and so on. Thus ML is about letting computers infer models of the world and ways of acting, as opposed to us telling them what to do precisely through excruciating programming. This course will also tackle some of the fundamental problems at the interface of learning, decision theory and probability. The course develops the theoretical foundations, representations and algorithms for active learning, value of information problems, experimental design, attention, optimal control and reinforcement learning. The course will present these developments in the the context of web crawling, relevance feedback in HCI, robotic exploration, question answering systems, clinical trials, active vision, active labelling of database entries, network problems, graphics and animation and optimal control. The pre-requisites are linear algebra, calculus and basic statistics or probability. I you don't have these skills, you have two weeks to acquire them before the course starts. I can provide background handouts if you come to my office early enough.