An evolving neural network to perform dynamic principal component analysis
Artikel i vetenskaplig tidskrift, 2009
Nonlinear principal component analysis is one of the best dimension reduction techniques developed during the recent years which have been applied in different signal-processing applications. In this paper, an evolving category of auto-associative neural network is presented which is applied to perform dynamic nonlinear principal component analysis. Training strategy of the network implements both constructive and destructive algorithms to extract dynamic principal components of speech database. In addition, the proposed network makes it possible to eliminate some dimensions of sequences that do not play important role in the quality of speech processing. Finally, the network is successfully applied to solve missing data problem.
Evolving auto-associative neural network
Dynamic principal component analysis
Missing data problem