A13: Deep Learning with HPC Simulations for Extracting Hidden Signals: Detecting Gravitational Waves
Author
Event Type
ACM Student Research Competition
Poster
TimeWednesday, November 15th3pm - 3:10pm
Location701
DescriptionWe introduce Deep Filtering, new machine learning method for end-to-end time-series signal processing, which combines two deep one-dimensional convolutional neural networks for classification and regression to detect and characterize signals much weaker than the background noise. We trained this method with a novel curriculum learning scheme on data derived from HPC simulations and applied it for gravitational wave analysis specifically for mergers of black holes and demonstrated that it significantly outperforms conventional machine learning techniques, is far more efficient than matched-filtering, offering several orders-of-magnitude speed-up, allowing real-time processing of raw big data with minimal resources, and extends the range of detectable signals. This initiates a new paradigm for scientific research which employs massively-parallel numerical simulations to train artificial intelligence algorithms that exploit emerging hardware architectures such as deep-learning-optimized GPUs. Our approach offers a unique framework to enable coincident detection campaigns of gravitational wave sources and their electromagnetic counterparts.
