• Corpus ID: 222066571

Few-shot Learning for Time-series Forecasting

@article{Iwata2020FewshotLF,
  title={Few-shot Learning for Time-series Forecasting},
  author={Tomoharu Iwata and Atsutoshi Kumagai},
  journal={ArXiv},
  year={2020},
  volume={abs/2009.14379}
}
Time-series forecasting is important for many applications. Forecasting models are usually trained using time-series data in a specific target task. However, sufficient data in the target task might be unavailable, which leads to performance degradation. In this paper, we propose a few-shot learning method that forecasts a future value of a time-series in a target task given a few time-series in the target task. Our model is trained using time-series data in multiple training tasks that are… 

Figures and Tables from this paper

Few-Shot Forecasting of Time-Series with Heterogeneous Channels

TLDR
This paper formalizes the problem of few-shot forecasting of time-series with heterogeneous channels for the first time and develops a model composed of permutation-invariant deep set-blocks which incorporate a temporal embedding.

Meta-Forecasting by combining Global Deep Representations with Local Adaptation

TLDR
This work introduces a novel forecasting method, called Meta GlobalLocal Auto-Regression (Meta-GLAR), that adapts to each time series by learning in closed-form the mapping from the representations produced by a recurrent neural network (RNN) to one-step-ahead forecasts.

Multimodal Meta-Learning for Time Series Regression

TLDR
This paper proposes a method for conditioning parameters of the model through an auxiliary network that encodes global information of the time series to extract meta-features and shows empirically that this proposed meta-learning method learns TSR with few data fast and outperforms the baselines in 9 of 12 experiments.

Meta-Learning for Koopman Spectral Analysis with Short Time-series

TLDR
A meta-learning method for estimating embedding functions from unseen short time- series by exploiting knowledge learned from related but different time-series is proposed and achieves better performance in terms of eigenvalue estimation and future prediction than existing methods.

Long-Range Transformers for Dynamic Spatiotemporal Forecasting

TLDR
This paper addresses multivariate forecasting into a “spatiotemporal sequence" formulation where each Transformer input token represents the value of a single variable at a given time and Long-Range Transformers can then learn interactions between space, time, and value information jointly along this extended sequence.

FloodDAN: Unsupervised Flood Forecasting based on Adversarial Domain Adaptation

  • Ruizhi ZhouYa‐Jun Pan
  • Computer Science, Engineering
    2022 IEEE 5th International Conference on Big Data and Artificial Intelligence (BDAI)
  • 2022
TLDR
Experimental results on Tunxi and Changhua flood dataset show that the proposed FloodDAN can perform flood forecasting effectively with zero target domain supervision and is on par with supervised models that uses 450–500 hours of supervision.

A Simple Baseline for Adversarial Domain Adaptation-based Unsupervised Flood Forecasting

TLDR
Experimental results on Tunxi and Changhua flood dataset show that FloodDAN can perform flood forecasting effectively with zero target domain supervision, and the performance of the floodDAN is on par with supervised models that uses 450-500 hours of supervision.

Few-shot Time-Series Forecasting with Application for Vehicular Traffic Flow

  • Victor TranA. Panangadan
  • Computer Science
    2022 IEEE 23rd International Conference on Information Reuse and Integration for Data Science (IRI)
  • 2022
TLDR
Deep neural network architectures for few-shot forecasting using a Siamese twin network approach to learn a difference function between pairs of time-series, rather than directly forecasting based on historical data as seen in traditional forecasting models.

References

SHOWING 1-10 OF 58 REFERENCES

Meta-learning framework with applications to zero-shot time-series forecasting

TLDR
The empirical results emphasize the importance of the identified meta-learning mechanisms for successful zero-shot univariate forecasting, suggesting that it is viable to train a neural network on a source TS dataset and deploy it on a different target TS dataset without retraining, resulting in performance that is at least as good as that of state-of-practice univariate forecast models.

N-BEATS: Neural basis expansion analysis for interpretable time series forecasting

TLDR
The proposed deep neural architecture based on backward and forward residual links and a very deep stack of fully-connected layers has a number of desirable properties, being interpretable, applicable without modification to a wide array of target domains, and fast to train.

Meta-learning how to forecast time series

TLDR
A random forest is used to identify the best forecasting method using only time series features and is shown to yield accurate forecasts comparable to several benchmarks and other commonly used automated approaches of time series forecasting.

Cross-domain Meta-learning for Time-series Forecasting

Time-series Extreme Event Forecasting with Neural Networks at Uber

TLDR
A novel endto-end recurrent neural network architecture is proposed that outperforms the current state of the art event forecasting methods on Uber data and generalizes well to a public M3 dataset used for time-series forecasting competitions.

Meta-Learning for Few-Shot Time Series Classification

TLDR
This work uses a triplet-loss based learning procedure that does not require any constraints to be enforced on the number of classes for the few-shot TSC tasks, and is the first to use meta-learning based pre-training for TSC.

Meta-learning for time series forecasting and forecast combination

Few-shot Time-series Classification with Dual Interpretability

TLDR
Dual Prototypical Shapelet Networks (DPSN) is proposed, which not only train a neural networkbased model but also interpret the model from the perspective of representative time series samples and shapelets, and outperforms state-of-the-art time-series classification methods, especially when trained with few data.

Optimization as a Model for Few-Shot Learning

...