Correlation of internal representations in feed-forward neural networks

Andreas Engel

doi:10.1088/0305-4470/29/13/003

Home/Archive/arXiv

arXiv · cond-mat/9604070

Correlation of internal representations in feed-forward neural networks

Andreas Engel

View original source ↗Read source document ↗Export metadata ↓

Abstract

Feed-forward multilayer neural networks implementing random input-output mappings develop characteristic correlations between the activity of their hidden nodes which are important for the understanding of the storage and generalization performance of the network. It is shown how these correlations can be calculated from the joint probability distribution of the aligning fields at the hidden units for arbitrary decoder function between hidden layer and output. Explicit results are given for the parity-, and-, and committee-machines with arbitrary number of hidden nodes near saturation.

Explore related subjects

cond-mat

Keep this discovery

BibTeX RIS

Andreas Engel. 1996-04-11. Correlation of internal representations in feed-forward neural networks. https://doi.org/10.1088/0305-4470%2F29%2F13%2F003

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2015

Spectral Representation for the Effective Macroscopic Response of a Polycrystal: Application to Third-Order Nonlinear Susceptibility

Erratum: In our paper, we show that the spectral representation for isotropic two-component composites also applies to uniaxial polycrystals. We have learned that this result was, in fact, first conjectured by G.W. Milton. While our derivation is more detailed, our result for the spectral function is the same as Milton's. We very much regret not having been aware of this work at the time of writing our paper. Original abstract: We extend the spectral theory used for the calculation of the effective linear response functions of composites to the case of a polycrystalline material with uniaxially anisotropic microscopic symmetry. As an application, we combine these results with a nonlinear decoupling approximation as modified by Ma et al., to calculate the third-order nonlinear optical susceptibility of a uniaxial polycrystal, assuming that the effective dielectric function of the polycrystal can be calculated within the effective-medium approximation.

cond-mat↗

arXiv2014

Towards quantum well hot hole lasers

It should be possible to improve hot-hole laser performance by moving from bulk materials to a quantum well structure. The extra design parameters enable us to alter the band structure by changing the crystal orientation of the growth direction; to use the well width to shift the subband offsets, enabling the effect of the LO phonon scattering cut-off to be controlled; and to use modulation doping to ensure a high hole concentration to increase the gain without the dopants being present in the gain region. We present the first simulations of THz quantum well hot-hole lasers that can produce inversion and optical gain.

cond-mat↗

arXiv2014

The Hall Effect and ionized impurity scattering in Si$_{(1-x)}$Ge$_x$

Using Monte Carlo simulations, we demonstrate that including ionized impurity scattering in models of Si$_{(1-x)}$Ge$_x$ is vital in order to predict the correct Hall parameters. Our results show good agreement with the experimental data of Joelsson et.al. (JAP, 1997).

cond-mat↗