Lab Lunch: 27 November 2018 - James Cheney

Title: Foundational Anomaly Detection

Abstract:

Given some data, it is often of interest to separate it into "typical" and "unusual" parts, where the latter might be of greater interest for detecting errors, security vulnerabilities or attacks, or fraud.  There is a very large literature on anomaly detection or outlier detection, but much of it is ad hoc, problem-specific, or tailored to continuous/numerical data.  There is much less work on discrete/categorical anomaly detection and little on foundations.  In this talk I'll present the results of  preliminary attempts to provide some.