Generalized approach to matched filtering using neural networks

Gravitational wave science is a pioneering field with rapidly evolving data analysis methodology currently assimilating and inventing deep learning techniques. The bulk of the sophisticated flagship searches rely on time-tested matched filtering principle within their core. In this paper, we make key observation relationship between emerging traditional techniques: formally equivalent to particular neural network. This means that network can be constructed analytically exactly implement filtering, further trained or boosted additional complexity for improved performance. Moreover, show proposed architecture outperform both without knowledge prior parameter distribution. When given, approach statistically optimal We also propose investigate two different architectures MNet-Shallow MNet-Deep, which at initialization data. has simpler structure, while MNet-Deep more flexible deal wider range distributions. Our theoretical findings are corroborated by experiments using real LIGO synthetic injections, where our methods significantly false positive rates above $5\times 10^{-3}\%$. fundamental equivalence networks allows us define "complexity standard candle" characterize relative approaches gravitational signal in common framework. Finally, results suggest new perspectives role detection.