Tid: 8 september 2008 kl 15.15-16.00 .
Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!
Titel: Anomaly detection with Bayesian Statistics
Sammanfattning: At SICS we have recently been working a lot on statistical anomaly detection, i.e. to estimate a model over normal cases, and use that to detect abnormal cases by calculating their probability of being generated by the model.
By using Bayesian Statistics, it is possible to build models that are sensitive to anomalies already for small numbers of training samples, while at the same time robust against false alarms. Based on this we have constructed a method that combines anomaly detection, clustering, and classification.
I will describe the model, and give examples of successful cases where this method has been used.
|Sidansvarig: Filip Lindskog