Regularization as a Substitute for Pre-Processing of Data in Neural Network Training
1991 (English)Report (Other academic)
The great importance of pre-processing of data before the training of a feed forward network is emphasised by many researchers. This pre-processing is not always straightforward, and, further, the need of pre-processing makes the model “less black”. We show that regularization, besides its other positive effects, reduces the need of pre-processing.
Place, publisher, year, edition, pages
Linköping: Linköping University , 1991.
LiTH-ISY-I, ISSN 8765-4321 ; 1296
Neural nets, Regularization, Bad start values, Pre-processing
IdentifiersURN: urn:nbn:se:liu:diva-55482OAI: oai:DiVA.org:liu-55482DiVA: diva2:316122