The elusive origin of black holes in binaries

Post GitHub

The question

Can we establish the origin of black holes detected in binaries?

The strategy

I wrangle a large dataset from a suit of simulations for dynamically-formed black-hole binaries. I use the set to train a machine-learning classification model. I cross-validate the model and apply it to messy data from interferometers.

The final product

A model that predicts the binary's formation path and how many dynamical interactions were undertaken by the black holes with 85 and 99 percent accuracy, respectively.

A polynomial regression model for the loudness of massive black-hole binaries

Post GitHub

The question

Can we detect large black holes with future gravitational-wave experiments?

The strategy

I create and wrangle a dataset from a large suite of simulations of the signal-to-noise ratio (SNR) of intermediate mass black holes, using dedicated software and parallel computing. I train a polynomial regression model to predict the SNR using a variety of detector configurations. I validate the model on both test data and dedicated mock simulations.

The final product

A regression model that, given a detector network, predicts whether an intermediate-mass black hole will be observed with 99 percent accuracy.

PopFisher: gravitational-wave hyperparameter estimation made easy!

Post GitHub Paper

The question

Can we quickly and effectively estimate the parameters that describe distributions of supermassive black holes?

The strategy

I used Monte Carlo analysis through Python's $\texttt{emcee}$ and a Fisher-matrix formalism to quickly and reliably estimate the index of the power law usually used to model the distribution of supermassive black holes.

The final product

A code that can estimate the hyperparameters of distributions of supermassive black holes in a matter of seconds.

An A/B test for biased gravitational-wave signals

Post GitHub Paper I Paper II

The question

Can we reliably estimate the parameters describing binary black holes, when their signals are buried in other loud signals?

The strategy

I create a mock dataset of both resolved and unresolved binary black hole time-series signals. I set up a Markov-Chain Monte Carlo infrastructure to estimate the parameters of an individual signal in the presence of the mock data.

The final product

A (semi-)analytical A/B testing infrastructure that reliably detects anomalies (biases) in the parameter estimation of time-series signals in the presence of an unmodelled or poorly-modelled background.