Reproduce “Time Series Forecasting With Deep Learning: A Survey” 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
Bryan Lim, Stefan Zohren
Bryan Lim and Stefan Zohren from the University of Oxford survey the landscape of deep learning in time series forecasting, systematically categorizing various neural network architectures, multi-horizon prediction strategies, and the integration of hybrid models. The survey details how these methods process temporal information, address uncertainty, and contribute to enhanced decision support through interpretability and counterfactual analysis.
Key insights
- Deep neural networks provide a flexible, data-driven paradigm capable of learning complex temporal dynamics, reducing the need for manual feature engineering.
- Different deep learning architectures leverage distinct mechanisms—local patterns by CNNs, internal memory by RNNs, and dynamic weighting by Attention—to capture diverse temporal features.
- Hybrid models, which combine deep learning's flexibility with statistical rigor, offer superior performance by effectively embedding domain knowledge, addressing limitations of pure data-driven or traditional approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.