Neural network methods in analysing and modelling time varying processes

摘要: Statistical data analysis is applied in many fields order to gain understanding the complex behaviour of system or process under interest. For this goal, observations are collected from process, and models built an effort capture essential structure observed data. In applications, e.g. control pattern recognition, modeled time-dependent, thus modeling temporal context essential. thesis, neural network methods statistical especially sequence processing (TSP) considered. Neural networks a class models, applicable tasks exploration regression classification. suitable for TSP can model time dependent phenomena, typically by utilizing delay lines recurrent connections within network. Recurrent Self-Organizing Map (RSOM) unsupervised capable sequences. The application RSOM with local prediction presented. divide input sequences into clusters, estimated corresponding these clusters. case studies, series problems Prediction results gained show better performance than conventional Map. sequence, which useful presented tasks. As another application, optimizing Web cache proposed. caches store recently requested objects, shared clients. A caching policy decides objects removed when storage space full. proposed approach predicts value each object features extracted object. Only syntactic used, enables efficient estimation model. be optimized based on predicted values cost designed according objectives caching. study, different stages decisions made during building suggest that application.