Making Sense of Complex Data

Catalog description

In the study of complex systems, it is common to be faced with large datasets: hundreds or thousands of neurons in a brain, genes in a cell, people in a society, firms in an economy or texts in a corpus. The challenge is to use such data to summarize, predict, comprehend and control system function. Explores modern approaches to extracting useful insights from data that are high-dimensional, heterogeneous, noisy and nonlinear. Moves from summarizing key properties at the population level (statistics) to characterizing predictable lower-dimensional patterns (dimensionality reduction) to building generative computational models with the appropriate level of complexity (model selection). Encounters PCA-type linear projections, nonlinear manifold techniques, topic modeling, clustering methods, network statistics and more. Also explores more abstract foundations for how these methods work and when they fail, from information bottleneck theory to Bayesian model selection. Students hone their data skills by applying state-of-the-art open-source software to real-world datasets.