CS Forum: Leman Akoglu, CMU "Anomaly Mining: Detection and Beyond"

Anomaly Mining: Detection and Beyond

Leman Akoglu
Carnegie Mellon University CMU

Abstract:

Anomaly mining is a key unsupervised learning task, with numerous applications in finance, security, surveillance, etc. Despite its importance and extensive work on the topic, anomaly mining remains a challenging subject in part due to the tremendous variety of both the forms that anomalies can take and the settings in which they are to be identified. One of the main thrusts of my recent research has been in tackling these challenges in anomaly mining by building detection models that are suitable for different practically-relevant settings. 

In the first part of my talk, I will highlight some vignettes from my recent work on anomaly detection with a concrete focus on a novel setting that we have investigated recently—anomaly detection with privileged information*—which has the potential to shepherd the field towards promising new directions including a deeper investigation of expert-knowledge-augmented, resource-frugal, and early anomaly detection. The second part of the talk will focus on challenges (and opportunities) for anomaly mining beyond the classical detection problem. Specifically, I will discuss recent and ongoing work on (i) interpretable explanations for anomalies and (ii) human-in-the-loop detection.

* See http://www.andrew.cmu.edu/user/shubhras/spi/

CS forum is a seminar series arranged at the CS department. The talks are intended for presentations of postdoctoral level researchers and professors, both for visiting and CS-department-based researchers.