Modern gravi­ta­tional-wave obser­va­to­ries such as LIGO, Virgo, and KAGRA, along with future projects like LISA, are essen­tial tools for under­stand­ing the physi­cal processes occur­ring through­out our Universe. However, the complex­ity and sheer volume of their data make analy­sis compu­ta­tion­ally demand­ing. My PhD research aims to stream­line the analy­sis of data from these obser­va­to­ries by lever­ag­ing modern machine learn­ing techniques, includ­ing Simula­tion-Based Infer­ence (SBI).

​

Obser­va­tions of gravi­ta­tional waves took the physics world by storm follow­ing the first detec­tion in 2016. Since then, LIGO, Virgo, and KAGRA have allowed us to probe various areas of physics, includ­ing general relativ­ity, cosmol­ogy, and stellar astro­physics. These advances rely on the analy­sis and charac­ter­i­za­tion of signals measured by the obser­va­to­ries. However, process­ing these signals is compu­ta­tion­ally expen­sive and can become a signif­i­cant bottle­neck in the analy­sis pipeline.

Recent advances in machine learn­ing have given rise to a plethora of domain-specific appli­ca­tions. One such technique is Simula­tion-Based Infer­ence (SBI), which can accel­er­ate infer­ence for various physi­cal problems by several orders of magni­tude while maintain­ing high accuracy. SBI has already been success­fully applied to gravi­ta­tional-wave analy­sis by our group, proving the applic­a­bil­ity of the framework.

My PhD will focus on devel­op­ing the SBI frame­work to better handle real-world datasets, includ­ing missing data and other imper­fec­tions. We also aim to extend these methods to upcom­ing gravi­ta­tional-wave obser­va­to­ries such as LISA. Ultimately, we want to make SBI models more capable, flexi­ble, and easier to use without sacri­fic­ing perfor­mance or accuracy, provid­ing a founda­tion for more efficient gravi­ta­tional-wave analysis.