Embracing Uncertainty: The New Machine Learning
Chris Bishop · 2011
Summary form only given. Computers are based on logic, but must increasingly deal with real-world data that is full of uncertainty and ambiguity. Modern approaches to machine learning use probability theory to quantify and compute with this uncertainty, and have led to a proliferation in the applications of machine learning, ranging from recommendation systems to web search, and from spam filters to voice recognition. Most recently, the Kinect 3D full-body motion sensor, which has become the fastest-selling consumer electronics product in history, relies crucially on machine learning. Furthermore, the advent of widespread internet connectivity, with centralised data storage and processing, coupled with recently developed algorithms for computationally efficient probabilistic inference, will create many new opportunities for machine learning over the coming years. The talk will be illustrated with tutorial examples, demonstrations, and real-world case studies.