Reproduce “Localized Weather Prediction Using Kolmogorov-Arnold Network-Based Models and Deep RNNs” 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
Ange-Clement Akazan, Verlon Roel Mbingui, Gnankan Landry Regis N'guessan, Issa Karambal
A study by the Σηigmα Research Group at AIMS RIC benchmarked Kolmogorov-Arnold Networks and deep recurrent neural networks for localized daily weather prediction in tropical African cities, showing KANs achieved R² values exceeding 0.99 for temperature while traditional recurrent networks performed stronger for atmospheric pressure.
Key insights
- Kolmogorov-Arnold Networks (KANs) are exceptionally effective for temperature forecasting in tropical climates, achieving nearly perfect R² values.
- Recurrent Neural Networks (RNNs) remain highly competitive and perform better than KAN-based models for atmospheric pressure prediction.
- Temporal Kolmogorov-Arnold Networks (TKANs), particularly customized variants, show promise for precipitation forecasting, especially in lower-rainfall regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.