Spline Filters For End-to-End Deep Learning
Randall Balestriero, Romain Cosentino, Herve Glotin, Richard Baraniuk
Abstract
We propose to tackle the problem of end-to-end learning for raw waveform signals by introducing learnable continuous time-frequency atoms. The derivation of these filters is achieved by defining a functional space with a given smoothness order and boundary conditions. From this space, we derive the parametric analytical filters. Their differentiability property allows gradient-based optimization. As such, one can utilize any Deep Neural Network (DNN) with these filters. This enables us to tackle in a front-end fashion a large scale bird detection task based on the freefield1010 dataset known to contain key challenges, such as the dimensionality of the inputs data ($>100,000$) and the presence of additional noises: multiple sources and soundscapes.
BibTeX
@InProceedings{pmlr-v80-balestriero18a,
title = {Spline Filters For End-to-End Deep Learning},
author = {Balestriero, Randall and Cosentino, Romain and Glotin, Herve and Baraniuk, Richard},
booktitle = {Proceedings of the 35th International Conference on Machine Learning},
pages = {364--373},
year = {2018},
editor = {Dy, Jennifer and Krause, Andreas},
volume = {80},
series = {Proceedings of Machine Learning Research},
month = {10--15 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v80/balestriero18a/balestriero18a.pdf},
url = {https://proceedings.mlr.press/v80/balestriero18a.html},
abstract = {We propose to tackle the problem of end-to-end learning for raw waveform signals by introducing learnable continuous time-frequency atoms. The derivation of these filters is achieved by defining a functional space with a given smoothness order and boundary conditions. From this space, we derive the parametric analytical filters. Their differentiability property allows gradient-based optimization. As such, one can utilize any Deep Neural Network (DNN) with these filters. This enables us to tackle in a front-end fashion a large scale bird detection task based on the freefield1010 dataset known to contain key challenges, such as the dimensionality of the inputs data ($>100,000$) and the presence of additional noises: multiple sources and soundscapes.}
}