Giovani in un'ora - Ciclo di seminari - Terza parte
- Day - Time: 12 February 2020, h.11:30
- Place: Area della Ricerca CNR di Pisa - Room: C-29
Giulio Masetti - "Enhancing the definition and analysis of large stochastic models"Abstract: Evaluating non-functional properties, such as performance, availability and reliability, of computer and cyber-physical systems is a demanding task because of the ever increasing complexity of the world around us. Model-based analysis can address non-functional properties, and nowadays free and commercial tools offer the possibility of defining and analyzing large stochastic models through the composition of smaller ones, but for very large models the composition approaches adopted so far show intrinsic limitations from the performance point of view. Thanks to the Grants for Young Mobility, I spent one month at Urbana-Champaign, Illinois, studying the implementation of a new composition operator, called Dependency-Aware Replication (DARep), within the Mobius modeling environment. The relationship between the new operator and existing ones has been investigated, and a comparison of different composition approaches has been carried out. During this talk, I will briefly discuss the ideas behind DARep and present some examples of application.
Luca Pappalardo - "Is that your home? Assessing the accuracy of home detection algorithms on ground truth for different data sources"
Abstract: As more and more research is developed using Call Detail Records (CDRs), it has become essential to analyze specific formal properties of these "real-time" data sets. In particular, most studies reported in the literature begin by finding where the users in their data sets live, i.e., their home locations. Many of these works apply their own "idiosyncratic" algorithms of home location detection: some of them use the last antenna at night or the first antenna in the morning, others use the antenna with the highest number of users appearances, and yet others take more sophisticated methodologies. The most pressing issue with these studies so far has been validation. It is hence hard to obtain a large enough sample of subscribers for which complete histories are available and, at the same time, for which the actual home location is known. In this talk, we present the first validation ever of home location detection algorithms on individual ground truth data, i.e., histories of mobile phone users for which the actual home address is known. We will show that, when using CDRs to describe mobility histories, most of the existing algorithms achieve poor accuracy on the ground truth data. Conversely, using denser variants of CDRs such as eXtended Detail Records (XDRs) and Control Plane (CP) allows achieving significantly better accuracy while using a smaller portion of the users' mobility history.