Since November 2020, I am a PhD student working here at Utrecht on Gravitational Gaves Data Analysis. I am part of the large effort to detect and such intriguing signal from exotic objects such as black holes and neutron star. As member of the GRASP, I am part of the Virgo collaboration that maintains a GW detector in Italy. Together with LIGO and Kagra, Virgo forms a worldwide collaboration to make the interferometer work, to share the data from them and to extract as much physics as possible.

Of course the collaboration gathers many different expertise, and everyone can only do a little part of it. My work is mostly based on searches and waveform modelling

Searches

Searching for gravitational waves is a very hard problem: you are looking for a signal o(1e-21) buried in noise o(1e-18)! For this reason a huge number of methods (and scientific careers) have been developed. My work is part of this effort: together with my colleagues, I am developing a pipeline (or to adapt the existing pipelines) for detecting signals from precessing BBHs. Besides this, I am exploring methods for detecting anomalies into the noise of the instrument: they might be signals, glitches or just random fluctuations.

In this task, I try to use as many machine learning techniques as possible: they are super exciting and also useful. Can really make a difference

Waveforms

In order to extract as much as information possible from a detection of GW, a deep knowledge of the wf to detect is required. There is an intense work on this, trying to build accurate templates for data analysis. As this requires working with gr, it can be very time consuming. I am working to speed up the waveform generation, thus making the analysis more efficient. With the aid of ML I made a copy of some existing model, capturing the WFs features at a fraction of computational cost. The model is called mlgw and you can see it here.
Currently I am working on an expansion of it, in order to include precessing waveforms

Projects & Publications

• mlgw: this is my master thesis, where I built a ML model to generate WFs from a generic, spin aligned BBH. The model is released as a python package called mlgw, and it provides an (hopefully) easy to use interface for waveform generation. The ML model is much faster than the usual approaches and opens up several opportunities in data analysis.
  Currently I am working on including precession in the model, thus making it even more realistic
  A paper accompayining the model is published on Physical Review Letter D at this link (also on arxiv:2011.01958)
• mbank: a method for metric template placement in high dimensional BBH regions. The metric approximation to the match allows for a fast bank generation, making feasible to cover a large space such as the precessing one.
  The code is publicly realeased and ready to use and it is an important step forward on my work on precessing searches.
• ML_ODE: this is a tiny project, which implements a ML model to solve any Ordinary Differential Equation (ODE). Once the model is trained properly, the neural network provides a fast and reliable approximation to the solution. This might be useful in many cases. For more details, you can visit the GitHub page of the project.
• More will come soon... :)