Paris workshop on Bayesian Deep Learning for Cosmology and Time Domain Astrophysics

Deep Learning Techniques for Time Series Analysis in the context of Gravitational Waves Detection

poster_s1_4

Non programmé

2m

Buffon Amphitheater (APC laboratory, Université Paris Cité)

Poster + lightning talk Time Domain Astrophysics Lightning talks

Orateurs

M. Vlad-Andrei Basceanu (Institute of Space Science, Romania)Dr Laurentiu-Ioan Caramete (Institute of Space Science, Romania)

Description

Here we present the results of our tests involving different Deep Learning (DL) algorithms in order to detect Gravitational waves (GW) in time-domain data from Massive Black Hole Binaries (MBHB) mergers. We selected three different neural networks (Shallow Multilayer Perceptron, Deep Multilayer Perceptron and a Deep Convolutional Neural Network) which are trained with simulated GW signals and noise produced in-house. The dataset consists of GW signals with the component masses ratios (q) in the range of 1-1501, GW signals injected into Gaussian Noise in the same ratio range and Gaussian Noise. The whole dataset is split into 5 classes as follows: A (q = 1-300), B (q = 301-749), C (q = 750-1200), D (q = 1201-1501) and the fifth class representing just noise (N). The results have direct implications to future ground like Einstein Telescope or space-based GW observatories such as LISA.