Building and Optimising Surrogate Models of Black Hole Merger Remnant Properties Using Neural Networks

Numerical relativity surrogate models are among the most accurate tools for gravitational wave data analysis, and recent advances have incorporated neural networks to achieve significant speedups. In this talk, I will present a new version of the model NRSur7dq4Remnant which utilises neural network fitting, for the first time developing a systematic framework to optimise the neural network architecture, hyperparameters, and training dataset size. Our approach formalises network configuration choices which are often obfuscated by high dimensionality of hyperparameters, and physically motivates the amount of training data required for neural network surrogate fitting. The final model achieves accuracy comparable to, or slightly better than, the original surrogate, with significant speedups.