Differentiating Functions of the Jacobian with Respect to the Weights

Abstract

For many problems, the correct behavior of a model depends not only on its input-output mapping but also on properties of its Jacobian matrix, the matrix of partial derivatives of the model's outputs with respect to its in(cid:173) puts. We introduce the J-prop algorithm, an efficient general method for computing the exact partial derivatives of a variety of simple functions of the Jacobian of a model with respect to its free parameters. The algorithm applies to any parametrized feedforward model, including nonlinear re(cid:173) gression, multilayer perceptrons, and radial basis function networks.

Cite

Text

Flake and Pearlmutter. "Differentiating Functions of the Jacobian with Respect to the Weights." Neural Information Processing Systems, 1999.

Markdown

[Flake and Pearlmutter. "Differentiating Functions of the Jacobian with Respect to the Weights." Neural Information Processing Systems, 1999.](https://mlanthology.org/neurips/1999/flake1999neurips-differentiating/)

BibTeX

@inproceedings{flake1999neurips-differentiating,
  title     = {{Differentiating Functions of the Jacobian with Respect to the Weights}},
  author    = {Flake, Gary William and Pearlmutter, Barak A.},
  booktitle = {Neural Information Processing Systems},
  year      = {1999},
  pages     = {435-441},
  url       = {https://mlanthology.org/neurips/1999/flake1999neurips-differentiating/}
}