Machine learning methods for numerical estimates of transport coefficients of quark-gluon plasma and the classification of urban landscapes

Student: Ben Hind
Supervisors: Pavel Buividovich (UoL), Carlos Medel Vera (UoL)
Institution: University of Liverpool

In this collaborative PhD project led jointly by the Department of Mathematical Sciences and the School of Architecture you will apply Machine Learning Methods to two scientifically different, but methodologically close problems. 

You will learn the framework of lattice QCD simulations, which is currently the only method for reliable calculations in the realm of strong nuclear interactions. In particular, you’ll concentrate on the extraction of transport coefficients from lattice QCD data, which involves a numerically ill-defined numerical analytic continuation procedure. You’ll train a deep Convolutional Neural Network and/or a Restricted Boltzmann Machine on the data obtained for small-scale or nearly-classical systems to apply it to quark-gluon plasma in experimentally relevant range of parameters. 

With guidance and data provided by the experts at the School of Architecture at the University of Liverpool, you will re-deploy deep Convolutional Neural Networks (CNNs) to implement the classification of urban spaces into categories such as “private space”, “public space”, “recreational area” etc.  

In collaboration with our industrial partner Art Recognition AG (Zurich), you will also have the opportunity to explore machine learning methods for artwork classification, focusing in particular on anomaly detection algorithms.  

Algorithm development will be co-supervised by AI experts from the Department of Mathematics.