|AZZALINI DAVIDE||Cycle: XXXIV |
Section: Computer Science and Engineering
Tutor: PRADELLA MATTEO
Advisor: AMIGONI FRANCESCO Major Research topic
:Methods and techniques for detecting anomalies in behaviors of intelligent autonomous systemsAbstract:
The objective of my work is to design a methodology and a set of related techniques to address those situations in which the behaviors of some (autonomous) devices are observed and there is the necessity to verify that some high level constraints on such behaviors are met. In general, the above constraints can be formulated as probabilistic (graphical) models of the nominal behaviors of the devices that can be either provided as a top-down specification or learned bottom-up from past observations.
The aim of this thesis is to find a way to match and compare the observed behaviors of devices with the nominal ones.
A plethora of different methodologies already exist for learning probabilistic behavioral models, for example, in the field of probabilistic robotics, Markovian models are commonly used to represent a robot's behavior. A lot of work has also been done in expressing constraints and requirements over dynamical systems, such as in the field of verifiable robotics, in which, starting from temporal logic specifications, it is possible to build robot controllers which are inherently correct by design.
Possible applications of my research include quantifying how much the observed behavior of a system is far from the nominal one with the aim of detecting anomalies. This would contribute to enable the so-called Long Term Autonomy, i.e., enabling intelligent systems to perform autonomously in complex, real-world scenarios over extended time periods.