Long-Term Sequential Prediction Using Expert Advice
share
For the prediction with expert advice setting, we consider methods to construct forecasting algorithms that suffer loss not much more than of any expert in the pool. In contrast to the standard approach, we investigate the case of long-term interval forecasting of time series, that is, each expert issues a sequence of forecasts for a time interval ahead and the master algorithm combines these forecasts into one aggregated sequence of forecasts. Two new approaches for aggregating experts long-term interval predictions are presented. One is based on Vovk's aggregation algorithm and considers sliding experts, the other applies the approach of Mixing Past Posteriors method to the long-term prediction. The upper bounds for regret of these algorithms for adversarial case are obtained. We also present results of numerical experiments of time series long-term prediction.
READ FULL TEXT