Reproduce “A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yao Qin, Dongjin Song, Haifeng Chen, Wei Cheng, Guofei Jiang, Garrison Cottrell

A Dual-Stage Attention-Based Recurrent Neural Network (DA-RNN) is introduced for multivariate time series prediction, employing distinct attention mechanisms to select relevant input features and important past time steps. The model consistently outperforms existing methods on real-world datasets like SML 2010 and NASDAQ 100, achieving more accurate and robust forecasts while providing interpretability into feature and temporal importance.

Key insights

  • A dual-stage attention mechanism, inspired by human cognition, can effectively address both feature selection and long-term dependency capture in time series data.
  • Explicitly attending to individual driving series (input attention) significantly improves prediction accuracy and provides interpretability regarding feature importance.
  • Combining input attention with temporal attention yields synergistic benefits, enabling robust performance even with noisy or high-dimensional multivariate inputs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.