[Talks@IDSIA] Multi-step time series forecasting

WHEN Tuesday (Feb 19, tomorrow) h. 11.00, room 223 (close to the vending machine) TITLE : Bias and variance investigation for multi-step time series forecasting ABSTRACT : Multi-step time series forecasting plays an important role in several fields of science and engineering, such as economics, finance, meteorology and telecommunication. As in any estimation problem, the bias and variance decomposition of the forecasting error is a fundamental concept whose understanding sheds some light on the factors affecting the performance. In the context of multi-step forecasting, others factors such as the forecasting strategy and the forecasting horizon can have an impact on the bias and variance components. Using non-linear and linear simulated time series, we compare four different forecasting strategies by decomposing their respective squared error in noise, bias and variance components throughout the forecasting horizon. Experiments on real-world time series from two forecasting competitions are also presented. SPEAKER : Souhaib Ben Taieb is a PhD candidate from Université Libre de Bruxelles co-supervised by Gianluca Bontempi and Rob J. Hyndman. The topic of his PhD thesis deals with machine learning strategies for multi-step forecasting. He recently ranked among the top five competitors in the Kaggle load forecasting competition.