First experiences with a neural-net parser
FESSANT ; MIDENET
Type de document
RAPPORT
Langue
anglais
Auteur
FESSANT ; MIDENET
Résumé / Abstract
In this work we investigate 2 neural network based models for correction and imputation problems. Our task is to evaluate the benefits of using such models for imputation of missing or erroneous data in surveys. First section briefly presents the problem we are dealing with, next we describe data we are working with. We present the 2 different kinds of neural network models we are studying. They seem well suited for imputation and erroneous data correction. The first model is a self-organizing map (Kohonen 95) and the second is a recurrent layered neural network with back propagation-type learning algorithm (Gingras and Bengio 96). Experiments and results are detailed. Conclusions and remarks are given in the final section.
