The adaptive forecasting task for multivariate time series whose characteristics can abruptly change in time by nonpredictable way is solved. The neural network architecture with synapses-dynamic filter is introduced and learning algorithm that combines both high speed and filtering properties is proposed. The proposed filter-predictor is intended for using in technical analysis tasks connected with on-line stock indexes prediction.
adaptive prediction and filtering, adaptive multivariate nonstationary time series, learning
"Adaptyvnыi fyltr-predyktor mnohomernыkh sushchestvenno nestatsyonarnыkh vremennыkh riadov" ,
Information Processing Systems,