Hierarchical Recurrent Neural Networks for Time Series Forecasting
Hierarchical Recurrent Neural Networks for Time Series Forecasting
Elia Cohen, M.Sc. student at the department of Industrial Advisor: Dr. Noam Koenigstein
Abstract:
This thesis investigates the use of Hierarchical Recurrent Neural Networks (HRNNs) for time series forecasting, addressing challenges like missing data and high volatility in disaggregated components. The proposed HRNN architecture propagates information across hierarchical levels using informative priors on RNN parameters. Applied to Covid-19 and Consumer Price Index (CPI) datasets, the HRNN model demonstrated superior accuracy at lower hierarchy levels compared to traditional RNNs. Future work will explore extending HRNNs to other datasets, highlighting their potential to enhance hierarchical time series forecasting.
Bio:
Elia Cohen, M.Sc. student at the department of Industrial Engineering department with over six years of experience in the tech industry. Currently, I work as a senior data scientist, driven by a strong passion for leveraging data and science to solve complex problems.
