Convolutional Neural Networks for Spectroscopic Data Analysis

Tuesday 10 October 2023, 4:30 pm
PhD student
J. Acquarelli drs.
prof. dr. E. Marchiori, prof. dr. L.M.C. Buydens
dr. T.M. van Laarhoven, dr. J.J. Jansen

In this thesis, we described spectroscopic data and analyzed them with deep learning inspired methodologies. The goal was to design artificial neural network-based methodologies capable of dealing with known challenges of the analysis of spectroscopic data. The main challenges are: 1) a limited number of samples; 2) high number of features; 3) noisy data. The goals are then to be robust against many types of noise affecting the samples and providing means for interpretation while achieving an high classification accuracy. In particular, we focused on developing convolutional neural network-based methodologies for the classification and interpretation of spectroscopic data. There are many types of spectroscopic data having different properties, but all sharing spectral locality: values  of neighboring wavelengths or wavenumbers that are not too much dissimilar  from each other. We exploited these properties in our approach. 

Jacopo Acquarelli was born in Umbertide, Perugia, Italy, on August 25, 1988. In 2007, he started his Bachelor’s program on information engineering at University of Siena, Italy, from which he graduated in April 2010. Subsequently he continues with a computer science Master’s program still at University of Siena. In December 2014, he defended his Master’s thesis entitled "Discovering potential clinical profiles of multiple sclerosis from clinical and pathological free text data with constrained non- 
negative matrix factorization" which was the result of its internship at the Radboud University Nijmegen. In February 2015, Jacopo started working on applying deep learning to chemometric data as a Ph.D. student in collaboration with the Analytical Chemistry Group at Institute for Molecules and Materials and the Data Science Group at Institute for Computing and Information Sciences, Radboud University Nijmegen. The project involved developing machine learning methodologies for spectroscopic data inspired from deep learning under the supervision of Prof. Elena Marchiori, Prof. Lutgarde M.C. Buydens, Dr. Twan van Laarhoven, and Dr. Jeroen J. Jansen. The results of the work he carried out during his PhD. period are described in this thesis. Since March 2019, Jacopo started working for a software quality company based in Eindhoven called TIOBE.