Matteo Calafà

• Currently doing research at Aarhus University (Denmark), Department of Mechanical and Production Engineering.
• We propose a new type of physics-informed neural network to efficiently solve the linear elasticity equations through an ad-hoc complex-valued architecture.
• A first article is already available.

• Work carried out at Microstudio S.R.L. (Italy), department of Research and Development.
• Implementation of analytical and computer vision methods for the 3D reconstruction of torsion springs from a sequence of images.
• The IVM machine, recently announced to the public, provides a cutting-edge technology for the 3D measuring of torsion springs and bent wires of any shape.

Modeling and analysis of the FoCS-2 Swiss fountain atomic clock

• Master thesis project at METAS (Bern, Switzerland), Time and Frequency laboratory.
• FEM calculations of the background pressure in the FoCS-2 atomic clock.
• Modeling of collisions between caesium and background atoms and the resulting role to the background pressure frequency shift.
• Exploration of optimal designs to reduce the background pressure shift.

• Research carried out during the internship at Kamstrup A/S (Denmark), department of Quality and Sustainability.
• Investigation and FreeFem implementation of a new and efficient DG method to obtain accurate simulations of the ultrasound propagations inside the Kamstrup flow meters.
• The method can work with millions of degrees of freedom and it then allows to obtain an in-depth analysis of the dynamics inside the Kamstrup products.
• This work has been followed by an arXiv preprint.

• Project for the course of Advanced Programming for Scientific Computing.
• A C++ header library which exploits lifex and deal.II to solve differential problems using the high-order discontinuous Galerkin method with Dubiner basis.
• Applications to electrophysiology including the monodomain problem with different ionic models.
• This work has been followed by an arXiv preprint.

Solving PDEs with log-normal random field coefficients

• Research project in collaboration with the CSQI chair of EPFL (prof. Fabio Nobile).
• Development of a method to solve diffusion equations with stochastic diffusion parameters.
• Research of error bounds for both the stochastic and finite element discretizations.

• Project for the course of Deep Learning.
• Definition of an efficient and performant ConvNet for a Noise2Noise model.
• Manual implementation of a ConvNet with forward and backpropagation algorithms.

• Project for the course of Turbulence.
• Analysis of turbulence characteristics from a dataset of velocity measurements in a wind tunnel.
• Comparisons with the K41 theory.
• Study on the chaotic and fractal properties of the generalized Baker’s map.

• Project for the course of Stochastic Simulation.
• Estimation of the risk of a pollutant particle to enter inside a drinking well region.
• Phyton implementation of different stochastic methods and Fenics implementation of the finite element method.

Pollutant Transport in Urban Canyons

• Project for the Environmental Transport Phenomena course.
• ANSYS Fluent implementation of a CFD simulation to predict the pollutant transport in different urban settings.

• Project for the Machine Learning course, in collaboration with the LASTRO laboratory (EPFL laboratory of astrophysics, Dr. Michele Bianco).
• Implemented a CNN to predict the propagation of radiation in the cosmic epoch of radiation.

• Project for the Machine Learning course.
• Implemented a supervised method to predict the observation of the Higgs Boson from the CERN dataset.
• With an accuracy of 83.6%, our work ranked 24 over 293 teams (TOP 9%).

• Project for the course of Numerical Analysis for Partial Differential Equations.
• MATLAB implementation of a 2D high-order Discontinuous Galerkin method.
• Analysis of the method on the bidomain problem of the cardiac electrophysiology.
• Performed simulations of heart pseudo-realistic phenomena.