Machine Learning

A branch of Artificial Intelligence, this scientific discipline uses automated learning to perform tasks without knowing the necessary explicit instructions. We use ML algorithms in order construct mathematical models which, based on sample data (commonly known as training data), are able to make prediction on data that the model has never seen. Nowadays machine learning techniques are widespread in the world of physics (jet tagging, model building, pdf determination, ...) and beyond (email filtering, computer vision, ...). Currently my main topic of research as part of N3PDF consist on the application of ML techniques to the determination of the internal structure of the proton.

The internal structure of the proton

One very important ingredient for the high energy physics program of the Large Hadron Collider (LHC) at CERN is the determination of the parton content of the proton. The internal structure of the proton in terms of quarks and gluons is given by the Parton Distribution Functions.

In the NNPDF collaboration we use Machine Learning techniques in order to determinate the aforementioned PDFs.

Some of our work on PDF determination has been featured in the cover of the European Physical Journal C.

QCD corrections to Higgs physics

The Higgs boson is at the center of the physics program of the LHC at CERN. After its discovery in 2012, the effort of the phenomenological community (hep-ph) has been focused on the more precise determination of its properties.

I have worked on some of the most precise calculations for the determination of the corrections due to Quantum Chromodynamics (QCD) effects to Higgs boson production on the gluon fusion and Vector Boson Fusion (VBF) production channels.

Numerical Calculations, Monte Carlo methods

State-of-the-art computations in High Energy Physics (HEP) require computing very complex multi-dimensional integrals numerically, as the analytical result is often not known. Monte Carlo algorithms are generally the option of choice be it in HEP applications or elsewhere, as the error of such algorithms does not grow with the number of dimensions.

Our aim is to improve the efficiency of widely used methods in order to enable more physics in less time. Some examples are our implementation of the Vegas algorithm in tensorflow (Vegasflow), PDFFlow or the distributed computing tool pyHepGrid.

New Hardware in High Energy Physics

Very complicated calculations are also very costly in terms of computing resources. As a result physics at the frontier of knowledge require technologies at the frontier of software and hardware. I am also dedicated to adapt these very complicated calculations to hardware accelerators which have seen an amazing development during the last decades. With a focus on Graphical Processing Units (GPUs) with packages like Vegasflow or PDFFlow, we are also considering Field Programmable Gate Arrays (FPGAs) or, more recently, we are exploring the usage of quantum circuits as co-processor in order to accelerate calculation, such as the QPDF for PDF fitting using Variational Quantum Circuits.