A next-generation gamma-ray observatory powered by Machine Learning techniques
PT-CERN Call 2021/1
Instituto Superior Técnico (Universidade de Lisboa)
Laboratório de Instrumentação e Física Experimental de Partículas
The main goal of this PhD thesis is to develop a next-generation detector concept with Machine Learning (ML) integration, suitable for the ambitious future experiment: the Southern Wide-Field Gamma-ray Observatory aimed to survey the Southern hemisphere sky. The design of a high-performance, cost-effective water Cherenkov detector should be able to cope with the observatory needs, particularly the capability to identify shower muons, essential to ensure an excellent gamma/hadron discrimination. The detector concepts studied in this thesis are based on the placement of multiple light sensors placed at the bottom of the station to explore the signal asymmetries created by the particles crossing it. The complex spatial and time signal structures will be analysed by state-of-the-art Machine learning-based algorithms, allowing to significantly reduce the complexity of the detector and its height. Therefore, this innovative technique would lead to smaller detector units, allowing the experiment to be placed at an extremely high altitude and lowering the energy threshold. These next-generation outdoor detectors with integrated ML techniques pose several challenges ahead, which shall be addressed in this thesis: the complexity of the problem and necessity to perform feature engineering; use of simulations with incomplete accurate descriptions; resilience to spurious effects (backgrounds)... Moreover, the optimisation of the position of the light sensors and the WCD dimensions shall be optimised to maximise the performance and cost of the station. Furthermore, the developed ML methods and detector concept will be used on the shower analysis, particularly the hadronic interaction properties. Such will allow extracting further information from the shower improving the overall observatory sensitivity to astrophysical sources.