Last updated on August 11, 2022
PhD (2018) in Computer Science from Sapienza University of Rome.
Work experience in both commercial and research based work environments in a variety of multidisciplinary contexts, including high school teaching.
This work presents an ambitious neural network aimed at predicting the accuracy of magnitude estimates computed shortly after an earthquake occurs. Machine learning in seismology has been regaining momentum in the past few years due to the increasing availability of seismic datasets. But the complex and heterogeneous nature of earthquakes still represents the main obstacle to a wider adoption of machine learning. The available datasets often turn out not to be appropriate for the automatic training of neural networks. Despite being just a preliminary study, the results demonstrate that machine learning is a viable approach to the problem at hand.
Cyber-Physical Systems (CPSs) have become an intrinsic part of the 21st century world. Systems like Smart Grids, Transportation, and Healthcare help us run our lives and businesses smoothly, successfully and safely. Since malfunctions in these CPSs can have serious, expensive, sometimes fatal consequences, Simulation Based Formal Verification (SBFV) tools are vital to minimise the likelihood of errors occurring during the development process and beyond. Their applicability is supported by the increasingly widespread use of Model Based Design (MBD) tools. MBD enables the simulation of CPS models in order to check for their correct behaviour from the very initial design phase. The disadvantage is that SBFV for complex CPSs is an extremely time-consuming process, which typically requires several months of simulation. Current SBFV tools are aimed at accelerating the verification process by computing all the simulation scenarios in advance in such a way as to split and simulate them in parallel with the use of multiple simulators. Furthermore, they compute optimised simulation campaigns in order to simulate common prefixes of these scenarios only once, thus avoiding redundant simulation. Nevertheless, there are still limitations that prevent a more widespread adoption of SBFV tools. Firstly, current tools cannot optimise simulation campaigns from existing datasets with collected scenarios. Secondly, there are currently no methods to predict the time required to complete the SBFV process. The lack of ability to predict the length of the process makes scheduling verification activities highly problematic. This work is aimed at overcoming these limitations with the use of a data-intensive simulation campaign optimiser and an accurate execution time estimator for simulation campaigns.
Licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.