Networks with Learned Unit Response Functions

Abstract

Feedforward networks composed of units which compute a sigmoidal func(cid:173) tion of a weighted sum of their inputs have been much investigated. We tested the approximation and estimation capabilities of networks using functions more complex than sigmoids. Three classes of functions were tested: polynomials, rational functions, and flexible Fourier series. Un(cid:173) like sigmoids, these classes can fit non-monotonic functions. They were compared on three problems: prediction of Boston housing prices, the sunspot count, and robot arm inverse dynamics. The complex units at(cid:173) tained clearly superior performance on the robot arm problem, which is a highly non-monotonic, pure approximation problem. On the noisy and only mildly nonlinear Boston housing and sunspot problems, differences among the complex units were revealed; polynomials did poorly, whereas rationals and flexible Fourier series were comparable to sigmoids.

Cite

Text

Moody and Yarvin. "Networks with Learned Unit Response Functions." Neural Information Processing Systems, 1991.

Markdown

[Moody and Yarvin. "Networks with Learned Unit Response Functions." Neural Information Processing Systems, 1991.](https://mlanthology.org/neurips/1991/moody1991neurips-networks/)

BibTeX

@inproceedings{moody1991neurips-networks,
  title     = {{Networks with Learned Unit Response Functions}},
  author    = {Moody, John and Yarvin, Norman},
  booktitle = {Neural Information Processing Systems},
  year      = {1991},
  pages     = {1048-1055},
  url       = {https://mlanthology.org/neurips/1991/moody1991neurips-networks/}
}