Machine learning is revolutionizing the sciences, but most existing methods require large amounts of human-generated training data to succeed. In this talk, we will introduce the unsupervised clustering problem, which requires an algorithm to make predictions without training data. We will discuss some classical methods for clustering before introducing a couple of new approaches. Throughout, connections with graph theory, Fourier analysis and probability theory will be developed. We will also demonstrate
applications to image processing and remote sensing.
James M. Murphy is an assistant professor of mathematics at Tufts University. His research interests include theoretical machine learning and applied harmonic analysis. He works on problems in unsupervised and semi-supervised learning, high-dimensional probability theory, image and signal processing, graph theory and frame theory.