Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

How is the Performance of a Neural Network Ensemble influenced by Over-fitting and Ensemble Size - A Numerical Investigation

Schwappach, Cordula Sybille LU (2014) FYTK01 20141
Computational Biology and Biological Physics - Has been reorganised
Abstract
The goal of this investigation is to study the performance of two different aspects of neural network ensembles: one ensemble contains over-trained networks and the other very many members. The over-training of the networks within the first ensemble is achieved with a larger number of hidden neurons than the approximation task in question would require. The result of the investigation is that over-training indeed has the power to improve an ensemble’s performance. The second question, if many ensemble members improve an ensemble’s output, is answered negatively, unless the stability of the prediction of individual input samples has to be enhanced. In this case, it can be said that the larger an ensemble is, the more stable is the... (More)
The goal of this investigation is to study the performance of two different aspects of neural network ensembles: one ensemble contains over-trained networks and the other very many members. The over-training of the networks within the first ensemble is achieved with a larger number of hidden neurons than the approximation task in question would require. The result of the investigation is that over-training indeed has the power to improve an ensemble’s performance. The second question, if many ensemble members improve an ensemble’s output, is answered negatively, unless the stability of the prediction of individual input samples has to be enhanced. In this case, it can be said that the larger an ensemble is, the more stable is the ensemble’s prediction. (Less)
Please use this url to cite or link to this publication:
author
Schwappach, Cordula Sybille LU
supervisor
organization
course
FYTK01 20141
year
type
M2 - Bachelor Degree
subject
keywords
Neural Network Ensemble, Over-Training, Number of Members, Data Analysis
language
English
id
4465718
date added to LUP
2014-06-17 07:47:57
date last changed
2020-01-24 12:57:49
@misc{4465718,
  abstract     = {{The goal of this investigation is to study the performance of two different aspects of neural network ensembles: one ensemble contains over-trained networks and the other very many members. The over-training of the networks within the first ensemble is achieved with a larger number of hidden neurons than the approximation task in question would require. The result of the investigation is that over-training indeed has the power to improve an ensemble’s performance. The second question, if many ensemble members improve an ensemble’s output, is answered negatively, unless the stability of the prediction of individual input samples has to be enhanced. In this case, it can be said that the larger an ensemble is, the more stable is the ensemble’s prediction.}},
  author       = {{Schwappach, Cordula Sybille}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{How is the Performance of a Neural Network Ensemble influenced by Over-fitting and Ensemble Size - A Numerical Investigation}},
  year         = {{2014}},
}