Electrical Engineering-Systems Dept.

סמינר

Oded Argon

(M.Sc. student under the supervision of Prof. Yuval Shavitt, joint work with Udi Weinsberg)

on the subject:

Inferring Multi-Periodicities in Large-Scale Datasets

Many processes in our daily lives are periodic. We wake up every day around the same time, read our emails and browse through our favorite social network. Weekends usually have a different routine, but the weekend is a periodic event within the week itself. Other examples of periodicities are engineered processes; our computer might get a new IP periodically if our network provider uses DHCP.

While the fact that these processes are periodic is not surprising, inferring the true periodicity is a non trivial task, especially when facing inherent measurement “noise”.

We present two methods for assessing the periodicity of processes and inferring their periodical patterns. We first convert the sampled process into a canonical binary (-1,1) signal and use Power Spectral Density for inferring a single dominant period that exists in the signal representing the sampling process. This method is highly robust to noise, but is most useful for single-period processes. Thus, we present a novel method for detecting multiple periods that comprise a single process, using iterative relaxation of the time-domain autocorrelation function of the same canonical signal.

We evaluate these methods using extensive simulations that include two noise models, sampling noise and phase noise, and show their applicability on real Internet measurements of end-host availability and IP address alternations.