Abstract

The problem of forecasting the future values of time series arises in numerous scientific fields and commercial applications. One of the difficulties of applying classical probabilistic forecasting methods as well as recent neural network-based techniques to large scale, real world data sets is the distance between the parametric assumptions made in a model and the true distribution of the observed data. We propose a flexible method for probabilistic modeling with conditional quantile functions using monotonic regression splines. The shape of the spline is parameterized by a neural network whose parameters are learned by minimizing the continuous ranked probability score. Unlike methods based on parametric probability density functions and maximum likelihood estimation, the proposed method can flexibly adapt to different output distributions without manual intervention. We empirically demonstrate the effectiveness of the approach on synthetic and real-world data sets.

Biography

Konstantinos Benidis received the M.Eng. degree from the School of Electrical and Computer Engineering, National Technical University of Athens, Greece, in 2011, the M.Sc. degree on information and Communication Technologies from the Polytechnic University of Catalonia, Spain, in 2013, and the Ph.D. degree in electronic and computer engineering from The Hong Kong University of Science and Technology in 2018. He is currently holding an Applied Scientist position at Amazon Web Services - Artificial Intelligence Labs (AWS - AI Labs), Berlin. His research interests include time series forecasting, deep learning and financial engineering.