Application of neural networks with echo states to time series predictionSupervisor: prof. RNDr. Jiří Pospíchal, DrSc.
This page shows details on the project. The primary projects are displayed together with a list of sub-projects.
|Project description:||The goal of the project is a study of a modern approach to reccurent neural networks, which is exceptionally suitable for a prediction of time series. Neural network in this case contains a block of neurons with a recurrent architecture, which is randomly generated and its weight coefficients are fixed during the learning of the network. Our preliminary numerical results show, that by an application of an incremental learning we can achieve an adapted neural network with Echo states, which has then a substantially better generalization ability. The proposed project will be aimed at a study of assorted methods of incremental learning of these networks and their aplication to prediction of time series produced in finance, energy industry, etc.|
|Kind of project:||VEGA ()|
|Department:||Institute of Applied Informatics (FIIT)|
|Project status:||Successfully completed|
|Project start date :||01. 01. 2007|
|Project close date:||31. 12. 2009|
|Number of workers in the project:||11|
|Number of official workers in the project:||0|