Feature discovery by competitive learning
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Asymptotically optimal block quantization
IEEE Transactions on Information Theory
Optimal adaptive k-means algorithm with dynamic adjustment of learning rate
IEEE Transactions on Neural Networks
Hi-index | 0.02 |
An approach to time series prediction of the CATS benchmark (for competition on artificial time series) is presented, where we use Fourier bandpass filters and competitive associative nets (CAN2s). Since one of the difficulties of this prediction is that the given time series does not seem to involve sufficient number of data for obtaining the underlying dynamics of the time series to reproduce low frequency components, we apply the CAN2 only for learning high frequency components extracted via Fourier bandpass filters with trial parameter values of the upper and lower cutoff frequencies and the missing last value of the given time series. Supposing that the optimal values among the trial values will give the best prediction performance for high frequency components, we can identify such optimal values via a certain reasonable validation method, with which we predict the missing high frequency components, and then we obtain the missing data to be predicted via adding high and low frequency components.