ENRICHMENT OF INNER INFORMATION REPRESENTATIONS IN BI-DIRECTIONAL COMPUTING ARCHITECTURE FOR TIME SERIES PREDICTION

A bi-directional computing architecture for time series prediction, which computes signal transformations for forward-time direction (present - future) and backward-time direction (present - past) bi-directionally, has been applied to several prediction tasks, and it shows a better score than the conventional uni-directional technique. But its detailed mechanism for an improvement of predicting accuracy has not been clear yet. Then, in order to solve this problem, the model's responses are investigated based on the principal component analysis approach in this paper. As a result, it is found that cooperation between the future and past prediction subsystems leads to enrichment of the inner information representations, and it gives the bi-directional model an advantage on signal processing abilities.