Anomaly Detection -- the art of finding "novel" pieces of data -- is a problem that has been attacked with a multitude of ad-hoc methods that are designed to sift through large amounts of data. Although these tools have shown utility for certain data mining applications, there has been no method available for assessing how well they actually "solve the anomaly detection problem" until recently. Researchers in CCS-3 made a fundamental discovery allowing them to develop the world's first practical method for estimating the performance of an anomaly detector. As a result, we can now compare detectors and answer the question, "Which of these systems is solving the anomaly detection problem most accurately?". Furthermore, this discovery has led to the development of a training algorithm that produces an anomaly detector with near-optimal results in polynomial time.

Selected Publications

D. Hush and J. Howse, Anomaly Detection on Graphs. Los Alamos National Laboratory Technical Report LA-UR-05-8440, 2005.   [   Abstract   |   Postscript (382 KB)   |   PDF (145 KB)   ]  

D. Hush, P. Kelly, C. Scovel and I. Steinwart, Provably Fast Algorithms for Anomaly Detection. Los Alamos National Laboratory Technical Report LA-UR-05-4367, 2005.   [   Abstract   |   Postscript (248 KB)   |   PDF (247 KB)   ]  

I. Steinwart, D. Hush, and C. Scovel, A classification framework for anomaly detection. Journal of Machine Learning Research, Vol. 6, pp. 211-232, 2005. Los Alamos National Laboratory Technical Report LA-UR-04-4716.   [   Abstract   |   PostScript (535 MB)   |   PDF (243 KB)   ]  

C. Scovel, D. Hush, C. Scovel and I. Steinwart, Learning Rates for Density Level Detection. Analysis and Applications, Vol. 3, No. 4 (2005) 356-371. Los Alamos National Laboratory Technical Report LA-UR-05-2088.   [   Abstract   |   Postscript (262 KB)   |   PDF (287 KB)   ]  

I. Steinwart, D. Hush, and C. Scovel, Density Level Detection is Classification. In Neural Information Processing Systems 17, pp. 1337-1344, (2005). Los Alamos National Laboratory Technical Report LA-UR-04-3768.   [   Abstract   |   PDF (104 KB)   ]