User login

Weekly Report - 11/08/13




Spent a couple of days at the start of the week reading papers which compared different anomaly detection methods in sequential time series data. Found some methods which might be worth looking into (Hidden Markov Models, Finite State Automata, and some window-based methods(kNN, STIDE, etc)). Spent the next couple of days reading up on the HMM, the Baum-Welch algorithm that it uses and also a Genetic Algorithm (which is supposedly more efficient at estimating the parameters in HMMs than the Baum-Welch algorithm). The papers' results seem quite promising, but unfortunately none disclose their methods in detais and I can't figure out how to implement HMMs with NetEvMon yet. Did find a C++ and a Java implementation, so it might be worth going over those.

Plan for next week: write up a brief proposal of the Masters project, talk to someone from the Stats dept. for additional help on implementing a working version of a detector that uses HMM.