A hybrid method of exponential smoothing and recurrent neural networks for time series forecasting
By Slawek Smyl. 2020
764 citations recorded by the source corpus.
Abstract This paper presents the winning submission of the M4 forecasting competition. The submission utilizes a dynamic computational graph neural network system that enables a standard exponential smoothing model to be mixed with advanced long short term memory networks into a common framework. The result is a hybrid and hierarchical forecasting method.
Subfields: Electrical and Electronic Engineering, Management Science and Operations Research, Model Architectures, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Electrical and Electronic Engineering
- Pierre Pinson
- Thordis Linda Thorarinsdottir
- Fotios Petropoulos
- Christoph Bergmeir
- Ricardo J. Bessa
- Florian Ziel
- Cheonbok Park
- Paul Goodwin
In Management Science and Operations Research
- Fotios Petropoulos
- Rob J Hyndman
- Konstantinos Nikolopoulos
- Nikolaos Kourentzes
- Vassilios Assimakopoulos
- Spyros Makridakis
- Christoph Bergmeir
- Evangelos Spiliotis
In Model Architectures
- Bryan Lim
- Bernie Wang
- Stefan Zohren
- Jan Gasthaus
- David Salinas
- Laurent Callot
- Michael Bohlke-Schneider
- Tim Januschowski
In Time-Series & Forecasting
- Fotios Petropoulos
- Bryan Lim
- Konstantinos Nikolopoulos
- Bernie Wang
- Jan Gasthaus
- Tim Januschowski
- Stefan Zohren
- Rob J Hyndman
Citations in the corpus
- Time-series forecasting with deep learning: a survey (2020, methodology): “e, hybrid models allow for the separation of stationary and non-stationary components, and avoid the need for custom input pre-processing. An example of this is the Exponential Smoothing RNN (ES-RNN) [64], winner of the M4 competition, which uses exponential smoothing to capture non-stationary trend”
- N-BEATS: Neural basis expansion analysis for interpretable time series forecasting (2019, background): “Finally, DL/TS hybrid is the winner of M4 competition (Smyl, 2020).”
- Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift (2022, methodology): “In the case of ES-RNN ∗ , the training of the model is unstable, failing to converge for several cases.”
- Deep Learning for Time Series Forecasting: Tutorial and Literature Survey (2020, methodology): “A further example in the global-local model class is the ES-RNN model proposed by Smyl [169] that has recently attracted attention by winning the M4 competition [129] by a large margin on both evaluation settings.”
- Forecasting: theory and practice (2020, methodology): “103 The first applications of NNs in Finance and currently the most widespread, is in financial trading.”
- Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting (2021, methodology): “Neural time series methods have recently become popular ways of solving the prediction problem via univariate point forecasting methods (Oreshkin et al., 2020; Smyl, 2020) or univariate probabilistic methods (Salinas et al.”
- Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions (2019, background): “Most notably, in the recent M4 competition, a Recurrent Neural Network (RNN) was able to achieve impressive performance and win the competition (Smyl, 2020).”