ICML 2021spotlight38 citations

Tractable structured natural-gradient descent using local parameterizations

Wu Lin, Frank Nielsen, Khan Mohammad Emtiyaz, Mark Schmidt

Abstract

Natural-gradient descent (NGD) on structured parameter spaces (e.g., low-rank covariances) is computationally challenging due to difficult Fisher-matrix computations. We address this issue by using \emph{local-parameter coordinates} to obtain a flexible and efficient NGD method that works well for a wide-variety of structured parameterizations. We show four applications where our method (1) generalizes the exponential natural evolutionary strategy, (2) recovers existing Newton-like algorithms, (3) yields new structured second-order algorithms, and (4) gives new algorithms to learn covariances of Gaussian and Wishart-based distributions. We show results on a range of problems from deep learning, variational inference, and evolution strategies. Our work opens a new direction for scalable structured geometric methods.

BibTeX
@InProceedings{pmlr-v139-lin21e,
  title = 	 {Tractable structured natural-gradient descent using local parameterizations},
  author =       {Lin, Wu and Nielsen, Frank and Emtiyaz, Khan Mohammad and Schmidt, Mark},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  pages = 	 {6680--6691},
  year = 	 {2021},
  editor = 	 {Meila, Marina and Zhang, Tong},
  volume = 	 {139},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {18--24 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v139/lin21e/lin21e.pdf},
  url = 	 {https://proceedings.mlr.press/v139/lin21e.html},
  abstract = 	 {Natural-gradient descent (NGD) on structured parameter spaces (e.g., low-rank covariances) is computationally challenging due to difficult Fisher-matrix computations. We address this issue by using \emph{local-parameter coordinates} to obtain a flexible and efficient NGD method that works well for a wide-variety of structured parameterizations. We show four applications where our method (1) generalizes the exponential natural evolutionary strategy, (2) recovers existing Newton-like algorithms, (3) yields new structured second-order algorithms, and (4) gives new algorithms to learn covariances of Gaussian and Wishart-based distributions. We show results on a range of problems from deep learning, variational inference, and evolution strategies. Our work opens a new direction for scalable structured geometric methods.}
}