N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
By Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, Yoshua Bengio. International Conference on Learning Representations, 2019
1,690 citations recorded by the source corpus.
We focus on solving the univariate times series point forecasting problem using deep learning. We propose a deep neural architecture based on backward and forward residual links and a very deep stack of fully-connected layers. The architecture has a number of desirable properties, being interpretable, applicable without modification to a wide array of target domains, and fast to train. We test the proposed architecture on several well-known datasets, including M3, M4 and TOURISM competition datasets containing time series from diverse domains. We demonstrate state-of-the-art performance for two configurations of N-BEATS for all the datasets, improving forecast accuracy by 11% over a statistical benchmark and by 3% over last year's winner of the M4 competition, a domain-adjusted hand-crafted hybrid between neural network and statistical time series models. The first configuration of our model does not employ any time-series-specific components and its performance on heterogeneous datasets strongly suggests that, contrarily to received wisdom, deep learning primitives such as residual blocks are by themselves sufficient to solve a wide range of forecasting problems. Finally, we demonstrate how the proposed architecture can be augmented to provide outputs that are interpretable without considerable loss in accuracy.
Subfields: Management Science and Operations Research, Signal Processing, Time-Series & Forecasting
View Paper: https://arxiv.org/abs/1905.10437
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Rob J Hyndman
- Fotios Petropoulos
- Konstantinos Nikolopoulos
- Nikolaos Kourentzes
- Vassilios Assimakopoulos
- Spyros Makridakis
- Christoph Bergmeir
- Evangelos Spiliotis
In Signal Processing
- Qiang Xu
- Ailing Zeng
- Haixu Wu
- Rob J Hyndman
- Mingsheng Long
- Bernie Wang
- Jianmin Wang
- Vassilios Assimakopoulos
In Time-Series & Forecasting
- Fotios Petropoulos
- Rob J Hyndman
- Konstantinos Nikolopoulos
- Bernie Wang
- Christoph Bergmeir
- Jan Gasthaus
- Tim Januschowski
- Spyros Makridakis
Citations in the corpus
- SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction (2021, methodology): “With the increasing data availability and computing power in recent years, it is shown that deep learning-based TSF techniques have the potential to achieve better forecasting accuracy than conventional approaches [24, 29].”
- 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., 2019b).”
- CoST: Contrastive Learning of Disentangled Seasonal-Trend Representations for Time Series Forecasting (2022, background): “Oreshkin et al. (2020) proposed a univariate deep forecasting model and showed that deep models outperform classical time series forecasting techniques.”
- TSMixer: An all-MLP Architecture for Time Series Forecasting (2023, neutral): “, 1970) N-BEATS (Oreshkin et al., 2020) LTSF-Linear (Zeng et al.”
- TS2Vec: Towards Universal Representation of Time Series (2021, methodology): “For forecasting tasks, the results of TCN, StemGNN and N-BEATS for all datasets and all baselines for Electricity dataset are based on our reproduction.”
- Deep Learning for Forecasting (2020, methodology): “Then, we will see how deep learning can be applied to forecasting by reviewing several state-of-the-art neural forecasting models (e.g., WaveNet [8], DeepAR [6], NBEATS [3] and the sequence-to-sequence model family [7, 10]).”
- ETSformer: Exponential Smoothing Transformers for Time-series Forecasting (2022, positive): “Univariate baselines further include N-BEATS (Oreshkin et al., 2019), DeepAR (Salinas et al.”
- FITS: Modeling Time Series with 10k Parameters (2023, methodology): “We report the comparison with NBeats (Oreshkin et al., 2019), NHits (Challu et al., 2023) and other transformer-based methods in the appendix.”
- Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift (2022, methodology): “Despite the simplicity of our method, there have been no cases of using such techniques in modern deep-learning-based time-series forecasting approaches (Zhou et al., 2021; Liu et al., 2021; Oreshkin et al., 2020).”
- Forecasting: theory and practice (2020, methodology): “Second, methods trained in a particular dataset can be effectively used to provide forecasts for series of different datasets that display similar characteristics (transfer-learning), thus allowing the development of generalised forecasting methods (Oreshkin et al., 2020a).”
- Decoupled Dynamic Spatial-Temporal Graph Neural Network for Traffic Forecasting (2022, neutral): “short- and long-term dependencies [17, 25, 37].”
- Non-stationary Transformers: Exploring the Stationarity in Time Series Forecasting (2022, methodology): “N-BEATS N-BEATS [27] proposes the explicit decomposition of trend and seasonal terms with strong interpretability.”